Information processing method, device and equipment

Abstract

The present invention provides an information processing method, device and equipment, wherein the information processing method includes: determining the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the time series feature category information of the network indicators; determining the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order; obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current value of other hyperparameters; updating the current value of the current hyperparameter to be adjusted according to the initial target value; wherein the other hyperparameters are the parameters of the hyperparameters of the network indicators except the current hyperparameter to be adjusted. This solution can solve the problems of large workload, waste of manpower, inability to realize automation and poor adjustment effect in the information processing scheme for hyperparameter adjustment of the Prophet algorithm for network indicators in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular to an information processing method, device and equipment. Background Art

[0002] At present, there are a large number of network performance indicators in mobile communication networks that need to be predicted and analyzed. Most of these indicators are time series indicators. The Prophet algorithm involved is based on time series decomposition, and uses full Bayesian or maximum a posteriori probability estimation algorithms to fit the time series components of each part, and comprehensively predicts the prediction results of each part. The algorithm takes into account factors such as multiple cycle superposition, trend, and holiday impact. The hyperparameters can be adjusted based on the experience of operation and maintenance experts, which is very suitable for predicting the trend of operation and maintenance indicators in mobile communication networks. At the same time, the algorithm also has the following advantages: it can fit time series with trend characteristics; no feature engineering is required; the hyperparameters are highly interpretable and can be adjusted based on expert experience; the algorithm generates statistical upper and lower bounds, which can be used to predict anomalies. Therefore, it is favored in existing network applications.

[0003] However, the Prophet algorithm has a large number of hyperparameters. In addition to the underlying algorithm full Bayesian and maximum a posteriori probability estimation related hyperparameters, it also has a large number of high-level hyperparameters open to operation and maintenance experts to adjust. If the parameters are adjusted manually, on the one hand, it is necessary to be very familiar with the technical principles of the algorithm, which is not realistic for network operation and maintenance experts; on the other hand, due to the different time series characteristics of different network indicators, targeted parameter adjustment is required. For network elements with more than 100 network indicators, the workload of parameter adjustment is huge. Therefore, in the current application of the algorithm, due to the lack of professional personnel or tools for parameter adjustment, the algorithm cannot achieve good performance (for some network indicators, the performance gap before and after parameter adjustment is close to 40%), which affects the application and promotion of the algorithm.

[0004] In addition, from the perspective of parameter adjustment technology, since the hyperparameters of the Prophet algorithm are mostly parameters that describe the characteristics of the time series after disassembly, they are very different from the hyperparameters of neural networks in deep learning. The parameter adjustment method of deep learning cannot be directly adopted, and it is necessary to design it according to the characteristics of the algorithm.

[0005] As can be seen from the above, the information processing scheme for hyper-parameter adjustment of the Prophet algorithm for network indicators in the prior art has problems such as large workload, waste of manpower, inability to achieve automation, and poor adjustment effect. Summary of the invention

[0006] The purpose of the present invention is to provide an information processing method, device and equipment to solve the problems in the prior art of the information processing scheme for hyper-parameter adjustment of the Prophet algorithm for network indicators, such as large workload, waste of manpower, inability to achieve automation and poor adjustment effect.

[0007] In order to solve the above technical problems, an embodiment of the present invention provides an information processing method, including:

[0008] Determine, according to the time series feature category information of the network indicator, the parameter adjustment order of the hyperparameters of the network indicator and the candidate values ​​of each of the hyperparameters;

[0009] According to the parameter adjustment order, determine the current hyperparameter to be adjusted from each of the hyperparameters;

[0010] Obtaining an initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters;

[0011] Updating the current value of the current hyperparameter to be adjusted according to the initial target value;

[0012] The other hyperparameters are the hyperparameters of the network indicators except the hyperparameters to be adjusted currently.

[0013] Optionally, obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters includes:

[0014] According to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters, obtaining the effect parameter value of the network indicator corresponding to each candidate value of the current hyperparameter to be adjusted;

[0015] According to the effect parameter value, an initial target value of the current hyperparameter to be adjusted is determined from various candidate values ​​of the current hyperparameter to be adjusted.

[0016] Optionally, after updating the current value of the hyperparameter to be adjusted according to the initial target value, the method further includes:

[0017] Obtaining the impact amplitude value corresponding to each of the hyperparameters;

[0018] According to the impact amplitude value, updating the parameter adjustment sequence;

[0019] According to the updated parameter adjustment order, return to execute the parameter adjustment order, determine the current hyperparameter to be adjusted from each hyperparameter; obtain the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; update the current value of the current hyperparameter to be adjusted according to the initial target value.

[0020] Optionally, obtaining the influence amplitude value corresponding to each of the hyperparameters includes:

[0021] Obtaining an influence amplitude value corresponding to the current hyperparameter to be adjusted according to an effect parameter value corresponding to the current value of the hyperparameter to be adjusted before being updated and an effect parameter value corresponding to the current value of the hyperparameter to be adjusted after being updated;

[0022] According to the influence amplitude value corresponding to the current hyperparameter to be adjusted, the influence amplitude value corresponding to each hyperparameter is obtained.

[0023] Optionally, before determining the adjustment order of the hyperparameters of the network indicators and the candidate values ​​of the hyperparameters according to the time series feature category information of the network indicators, the method further includes:

[0024] Determining the timing characteristic category information of the network indicator according to the timing characteristic parameter information of the network indicator;

[0025] The time series characteristic parameter information includes at least one of indicator direction information, time series characteristic information, value type information and value magnitude information.

[0026] Optionally, after updating the current value of the hyperparameter to be adjusted according to the initial target value, the method further includes:

[0027] Determine a group of first sampling points according to the initial target value, the adjustment step size and the number of first sampling points; the first sampling points are located on both sides of the initial target value;

[0028] Obtaining the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value;

[0029] According to the first optimal effect parameter value and the first worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

[0030] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value includes:

[0031] When the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is less than or equal to the first threshold, the adjustment step is increased according to a preset amplitude, and the step of determining a group of first sampling points according to the initial target value and the adjustment step is returned according to the increased adjustment step; until the adjustment step reaches an upper limit of the step, a target effect parameter value is obtained from each effect parameter value corresponding to each of the first sampling points, and the value of the first sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0032] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value includes:

[0033] When the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is greater than a first threshold, determining a search step length, a search direction, and a search endpoint; wherein the search step length is equal to the adjustment step length, and the search endpoint is a first sampling point at the end of the search direction;

[0034] According to the search step size, search direction and search endpoint, a final target value of the current hyperparameter to be adjusted is determined.

[0035] Optionally, determine the search direction, including:

[0036] A search direction is determined according to the initial target value and the first sampling point corresponding to the optimal effect parameter value.

[0037] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the search step size, the search direction, and the search endpoint includes:

[0038] Determine a set of second sampling points according to the search step size, the search direction, the search endpoint, and the number of second sampling points;

[0039] Obtaining the effect parameter value corresponding to each of the second sampling points to obtain a second optimal effect parameter value and a second worst effect parameter value;

[0040] According to the second optimal effect parameter value and the second worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

[0041] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0042] When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to the third threshold, re-determine the search step size, search direction and search endpoint;

[0043] According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint;

[0044] Among them, the re-determined search direction is consistent with the last search direction;

[0045] The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value;

[0046] The search endpoint is a second sampling point located at the end of the search direction;

[0047] The second threshold is smaller than the first threshold.

[0048] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0049] When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, determine that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted;

[0050] The second threshold is smaller than the first threshold.

[0051] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0052] When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0053] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0054] When the absolute value of the difference between the second best effect parameter value and the second worst effect parameter value is greater than a second threshold, re-determining the search direction;

[0055] Determining a final target value of the current hyperparameter to be adjusted according to the re-determined search direction;

[0056] The re-determined search direction is the direction from the second sampling point corresponding to the second worst effect parameter value to the second sampling point corresponding to the second best effect parameter value;

[0057] The second threshold is smaller than the first threshold.

[0058] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes:

[0059] When the re-determined search direction is consistent with the last search direction, or the re-determined search direction is not consistent with the last search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to a fourth threshold, re-determine the search step size and the search endpoint;

[0060] According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint;

[0061] The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value;

[0062] The search endpoint is a second sampling point located at the end of the search direction;

[0063] The second threshold is smaller than the first threshold.

[0064] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes:

[0065] When the re-determined search direction is inconsistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the fourth threshold, the value of the sampling point corresponding to the second optimal effect parameter value is determined to be the final target value of the current hyperparameter to be adjusted.

[0066] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes:

[0067] When the re-determined search direction is consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0068] The embodiment of the present invention also provides an information processing method, including:

[0069] Determine a set of first sampling points according to the initial target value of the hyperparameter to be adjusted, the adjustment step size, and the number of first sampling points; the first sampling points are located on both sides of the initial target value;

[0070] Obtaining the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value;

[0071] According to the first optimal effect parameter value and the first worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

[0072] An embodiment of the present invention further provides an information processing device, including:

[0073] A first determination module, used to determine the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the time series feature category information of the network indicators;

[0074] A second determination module is used to determine the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order;

[0075] A first processing module, configured to obtain an initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters;

[0076] A first updating module, used for updating a current value of the hyperparameter to be adjusted according to the initial target value;

[0077] The other hyperparameters are the hyperparameters of the network indicators except the hyperparameters to be adjusted currently.

[0078] Optionally, the first processing module includes:

[0079] A first processing submodule, configured to obtain, according to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters, an effect parameter value of the network indicator corresponding to each candidate value of the current hyperparameter to be adjusted;

[0080] The first determination submodule is used to determine the initial target value of the current hyperparameter to be adjusted from various candidate values ​​of the current hyperparameter to be adjusted according to the effect parameter value.

[0081] Optionally, also include:

[0082] A first acquisition module, configured to acquire the impact amplitude value corresponding to each of the hyperparameters after updating the current value of the hyperparameter to be adjusted according to the initial target value;

[0083] A second updating module, used for updating the parameter adjustment sequence according to the impact amplitude value;

[0084] The second processing module is used to return to execute the step of determining the current hyperparameter to be adjusted from the various hyperparameters according to the parameter adjustment order after the update; obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; and updating the current value of the current hyperparameter to be adjusted according to the initial target value.

[0085] Optionally, the first acquisition module includes:

[0086] A second processing submodule is used to obtain the influence amplitude value corresponding to the current hyperparameter to be adjusted according to the effect parameter value corresponding to the current value of the hyperparameter to be adjusted before the current value is updated and the effect parameter value corresponding to the current value of the hyperparameter to be adjusted after the current value is updated;

[0087] The third processing submodule is used to obtain the influence amplitude value corresponding to each of the hyperparameters according to the influence amplitude value corresponding to the current hyperparameter to be adjusted.

[0088] Optionally, also include:

[0089] A third determination module is used to determine the timing characteristic category information of the network indicator according to the timing characteristic parameter information of the network indicator before determining the parameter adjustment order of the hyperparameters of the network indicator and the candidate values ​​of each of the hyperparameters according to the timing characteristic category information of the network indicator;

[0090] The time series characteristic parameter information includes at least one of indicator direction information, time series characteristic information, value type information and value magnitude information.

[0091] Optionally, also include:

[0092] A fourth determination module is used to determine a set of first sampling points according to the initial target value, the adjustment step size and the number of first sampling points after updating the current value of the current hyperparameter to be adjusted according to the initial target value; the first sampling points are located on both sides of the initial target value;

[0093] A second acquisition module is used to acquire the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value;

[0094] The fifth determination module is used to determine the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value.

[0095] Optionally, the fifth determining module includes:

[0096] The fourth processing submodule is used to increase the adjustment step according to a preset amplitude when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is less than or equal to the first threshold value, and return to execute the step of determining a group of first sampling points according to the initial target value and the adjustment step according to the increased adjustment step; until the adjustment step reaches the step upper limit, the target effect parameter value is obtained from the effect parameter values ​​corresponding to the first sampling points, and the value of the first sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0097] Optionally, the fifth determining module includes:

[0098] A second determination submodule is used to determine a search step length, a search direction, and a search endpoint when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is greater than a first threshold; wherein the search step length is equal to the adjustment step length, and the search endpoint is a first sampling point located at the end of the search direction;

[0099] The third determination submodule is used to determine the final target value of the current hyperparameter to be adjusted according to the search step size, search direction and search endpoint.

[0100] Optionally, determine the search direction, including:

[0101] A search direction is determined according to the initial target value and the first sampling point corresponding to the optimal effect parameter value.

[0102] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the search step size, the search direction, and the search endpoint includes:

[0103] Determine a set of second sampling points according to the search step size, the search direction, the search endpoint, and the number of second sampling points;

[0104] Obtaining the effect parameter value corresponding to each of the second sampling points to obtain a second optimal effect parameter value and a second worst effect parameter value;

[0105] According to the second optimal effect parameter value and the second worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

[0106] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0107] When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to the third threshold, re-determine the search step size, search direction and search endpoint;

[0108] According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint;

[0109] Among them, the re-determined search direction is consistent with the last search direction;

[0110] The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value;

[0111] The search endpoint is a second sampling point located at the end of the search direction;

[0112] The second threshold is smaller than the first threshold.

[0113] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0114] When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, determine that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted;

[0115] The second threshold is smaller than the first threshold.

[0116] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0117] When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0118] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0119] When the absolute value of the difference between the second best effect parameter value and the second worst effect parameter value is greater than a second threshold, re-determining the search direction;

[0120] Determining a final target value of the current hyperparameter to be adjusted according to the re-determined search direction;

[0121] The re-determined search direction is the direction from the second sampling point corresponding to the second worst effect parameter value to the second sampling point corresponding to the second best effect parameter value;

[0122] The second threshold is smaller than the first threshold.

[0123] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes:

[0124] When the re-determined search direction is consistent with the last search direction, or the re-determined search direction is not consistent with the last search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to a fourth threshold, re-determine the search step size and the search endpoint;

[0125] According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint;

[0126] The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value;

[0127] The search endpoint is a second sampling point located at the end of the search direction;

[0128] The second threshold is smaller than the first threshold.

[0129] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes:

[0130] When the re-determined search direction is inconsistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the fourth threshold, the value of the sampling point corresponding to the second optimal effect parameter value is determined to be the final target value of the current hyperparameter to be adjusted.

[0131] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes:

[0132] When the re-determined search direction is consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0133] An embodiment of the present invention further provides an information processing device, including:

[0134] A sixth determination module, configured to determine a set of first sampling points according to the initial target value of the hyperparameter to be adjusted, the adjustment step size, and the number of first sampling points; the first sampling points are located on both sides of the initial target value;

[0135] A third acquisition module is used to acquire the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value;

[0136] The seventh determination module is used to determine the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value.

[0137] An embodiment of the present invention further provides an information processing device, including: a processor;

[0138] The processor is used to determine the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the time series feature category information of the network indicators;

[0139] According to the parameter adjustment order, determine the current hyperparameter to be adjusted from each of the hyperparameters;

[0140] Obtaining an initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters;

[0141] Updating the current value of the current hyperparameter to be adjusted according to the initial target value;

[0142] The other hyperparameters are the hyperparameters of the network indicators except the hyperparameters to be adjusted currently.

[0143] Optionally, obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters includes:

[0144] According to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters, obtaining the effect parameter value of the network indicator corresponding to each candidate value of the current hyperparameter to be adjusted;

[0145] According to the effect parameter value, an initial target value of the current hyperparameter to be adjusted is determined from various candidate values ​​of the current hyperparameter to be adjusted.

[0146] Optionally, the processor is further configured to:

[0147] After updating the current value of the hyperparameter to be adjusted according to the initial target value, obtaining the impact amplitude value corresponding to each of the hyperparameters;

[0148] According to the impact amplitude value, updating the parameter adjustment sequence;

[0149] According to the updated parameter adjustment order, return to execute the parameter adjustment order, determine the current hyperparameter to be adjusted from each hyperparameter; obtain the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; update the current value of the current hyperparameter to be adjusted according to the initial target value.

[0150] Optionally, obtaining the influence amplitude value corresponding to each of the hyperparameters includes:

[0151] Obtaining an influence amplitude value corresponding to the current hyperparameter to be adjusted according to an effect parameter value corresponding to the current value of the hyperparameter to be adjusted before being updated and an effect parameter value corresponding to the current value of the hyperparameter to be adjusted after being updated;

[0152] According to the influence amplitude value corresponding to the current hyperparameter to be adjusted, the influence amplitude value corresponding to each hyperparameter is obtained.

[0153] Optionally, the processor is further configured to:

[0154] Before determining the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the timing characteristic category information of the network indicators, determining the timing characteristic category information of the network indicators according to the timing characteristic parameter information of the network indicators;

[0155] The time series characteristic parameter information includes at least one of indicator direction information, time series characteristic information, value type information and value magnitude information.

[0156] Optionally, the processor is further configured to:

[0157] After updating the current value of the hyperparameter to be adjusted according to the initial target value, determining a set of first sampling points according to the initial target value, the adjustment step size, and the number of first sampling points; the first sampling points are located on both sides of the initial target value;

[0158] Obtaining the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value;

[0159] According to the first optimal effect parameter value and the first worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

[0160] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value includes:

[0161] When the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is less than or equal to the first threshold, the adjustment step is increased according to a preset amplitude, and the step of determining a group of first sampling points according to the initial target value and the adjustment step is returned according to the increased adjustment step; until the adjustment step reaches an upper limit of the step, a target effect parameter value is obtained from each effect parameter value corresponding to each of the first sampling points, and the value of the first sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0162] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value includes:

[0163] When the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is greater than a first threshold, determining a search step length, a search direction, and a search endpoint; wherein the search step length is equal to the adjustment step length, and the search endpoint is a first sampling point at the end of the search direction;

[0164] According to the search step size, search direction and search endpoint, a final target value of the current hyperparameter to be adjusted is determined.

[0165] Optionally, determine the search direction, including:

[0166] A search direction is determined according to the initial target value and the first sampling point corresponding to the optimal effect parameter value.

[0167] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the search step size, the search direction, and the search endpoint includes:

[0168] Determine a set of second sampling points according to the search step size, the search direction, the search endpoint, and the number of second sampling points;

[0169] Obtaining the effect parameter value corresponding to each of the second sampling points to obtain a second optimal effect parameter value and a second worst effect parameter value;

[0170] According to the second optimal effect parameter value and the second worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

[0171] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0172] When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to the third threshold, re-determine the search step size, search direction and search endpoint;

[0173] According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint;

[0174] Among them, the re-determined search direction is consistent with the last search direction;

[0175] The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value;

[0176] The search endpoint is a second sampling point located at the end of the search direction;

[0177] The second threshold is smaller than the first threshold.

[0178] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0179] When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, determine that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted;

[0180] The second threshold is smaller than the first threshold.

[0181] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0182] When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0183] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes:

[0184] When the absolute value of the difference between the second best effect parameter value and the second worst effect parameter value is greater than a second threshold, re-determining the search direction;

[0185] Determining a final target value of the current hyperparameter to be adjusted according to the re-determined search direction;

[0186] The re-determined search direction is the direction from the second sampling point corresponding to the second worst effect parameter value to the second sampling point corresponding to the second best effect parameter value;

[0187] The second threshold is smaller than the first threshold.

[0188] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes:

[0189] When the re-determined search direction is consistent with the last search direction, or the re-determined search direction is not consistent with the last search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to a fourth threshold, re-determine the search step size and the search endpoint;

[0190] According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint;

[0191] The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value;

[0192] The search endpoint is a second sampling point located at the end of the search direction;

[0193] The second threshold is smaller than the first threshold.

[0194] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes:

[0195] When the re-determined search direction is inconsistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the fourth threshold, the value of the sampling point corresponding to the second optimal effect parameter value is determined to be the final target value of the current hyperparameter to be adjusted.

[0196] Optionally, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes:

[0197] When the re-determined search direction is consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0198] An embodiment of the present invention further provides an information processing device, including: a processor;

[0199] The processor is used to determine a set of first sampling points according to the initial target value of the hyperparameter to be adjusted, the adjustment step size, and the number of first sampling points; the first sampling points are located on both sides of the initial target value;

[0200] Obtaining the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value;

[0201] According to the first optimal effect parameter value and the first worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

[0202] An embodiment of the present invention further provides an information processing device, comprising a memory, a processor, and a program stored in the memory and executable on the processor; when the processor executes the program, the above-mentioned information processing method is implemented.

[0203] An embodiment of the present invention further provides a readable storage medium on which a program is stored. When the program is executed by a processor, the steps in the above-mentioned information processing method are implemented.

[0204] The beneficial effects of the above technical solution of the present invention are as follows:

[0205] In the above scheme, the information processing method determines the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the timing feature category information of the network indicators; determines the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order; obtains the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current value of other hyperparameters; updates the current value of the current hyperparameter to be adjusted according to the initial target value; wherein the other hyperparameters are the parameters of the hyperparameters of the network indicators other than the current hyperparameter to be adjusted; it can realize automatic parameter adjustment for different timing indicators in the network, avoiding heavy workload and waste of manpower; and it can obtain better parameter configuration, thereby improving the performance of the algorithm and enhancing the adjustment effect; it also realizes automatic adjustment of hyperparameters (it can also realize automatic parameter adjustment based on the Prophet algorithm); it well solves the problems of heavy workload, waste of manpower, inability to realize automation, and poor adjustment effect in the information processing scheme for hyperparameter adjustment of the Prophet algorithm for network indicators in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0206] Figure 1 Schematic diagram of the information processing method provided by the embodiment of the present invention Figure 1 ;

[0207] Figure 2 A schematic diagram of a specific application flow of the information processing method provided by an embodiment of the present invention;

[0208] Figure 3 Schematic diagram of the information processing method provided by the embodiment of the present invention Figure 2 ;

[0209] Figure 4 Schematic diagram of the structure of the information processing device provided by the embodiment of the present invention Figure 1 ;

[0210] Figure 5 Schematic diagram of the structure of the information processing device provided by the embodiment of the present invention Figure 2 ;

[0211] Figure 6 Schematic diagram of the information processing device structure provided by the embodiment of the present invention Figure 1 ;

[0212] Figure 7 Schematic diagram of the structure of the information processing device provided by the embodiment of the present invention Figure 2 . DETAILED DESCRIPTION

[0213] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0214] The present invention aims at the problems that the information processing scheme for adjusting the hyper-parameters of the Prophet algorithm for network indicators in the prior art has a large workload, wastes manpower, cannot be automated, and has a poor adjustment effect. An information processing method is provided, including the following embodiments:

[0215] Embodiment 1

[0216] The information processing method provided by the embodiment of the present invention is as follows: Figure 1 As shown, including:

[0217] Step 11: Determine the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the time series feature category information of the network indicators;

[0218] Step 12: According to the parameter adjustment order, determine the current hyperparameter to be adjusted from each of the hyperparameters;

[0219] Step 13: Obtaining an initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters;

[0220] Step 14: updating the current value of the current hyperparameter to be adjusted according to the initial target value; wherein the other hyperparameters are the parameters of the hyperparameters of the network indicator except the current hyperparameter to be adjusted.

[0221] Among them, step 11 can be specifically implemented as determining the adjustment order of the hyperparameters for the network indicators and the candidate values ​​of each of the hyperparameters based on the timing feature category information of the network indicators and the pre-stored correspondence information; the pre-stored correspondence information may be the correspondence information between the category and the preset template, and the template includes the order and the candidate values.

[0222] The current values ​​of other hyperparameters in step 13 may be adjusted values ​​for the adjusted hyperparameters, and may be default values ​​for the unadjusted hyperparameters.

[0223] The information processing method provided in the embodiment of the present invention determines the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the timing feature category information of the network indicators; determines the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order; obtains the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current value of other hyperparameters; updates the current value of the current hyperparameter to be adjusted according to the initial target value; wherein the other hyperparameters are the parameters of the hyperparameters of the network indicators other than the current hyperparameter to be adjusted; it can realize automatic parameter adjustment for different timing indicators in the network, avoiding heavy workload and waste of manpower; and it can obtain better parameter configuration, thereby improving the performance of the algorithm and enhancing the adjustment effect; it also realizes automatic adjustment of hyperparameters (it can also realize automatic parameter adjustment based on the Prophet algorithm); it well solves the problems of heavy workload, waste of manpower, inability to realize automation and poor adjustment effect in the information processing scheme for hyperparameter adjustment of the Prophet algorithm for network indicators in the prior art.

[0224] Among them, obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters includes: obtaining the effect parameter value of the network indicator corresponding to each candidate value of the current hyperparameter to be adjusted according to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; and determining the initial target value of the current hyperparameter to be adjusted from each candidate value of the current hyperparameter to be adjusted according to the effect parameter value.

[0225] Furthermore, after updating the current value of the current hyperparameter to be adjusted according to the initial target value, it also includes: obtaining the influence amplitude value corresponding to each of the hyperparameters; updating the parameter adjustment order according to the influence amplitude value; returning to execute the step of determining the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order after the update; obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; and updating the current value of the current hyperparameter to be adjusted according to the initial target value.

[0226] Among them, the obtaining of the influence amplitude value corresponding to each of the hyperparameters includes: obtaining the influence amplitude value corresponding to the current hyperparameter to be adjusted according to the effect parameter value corresponding to the current value of the hyperparameter to be adjusted before being updated and the effect parameter value corresponding to the current value of the hyperparameter to be adjusted after being updated; obtaining the influence amplitude value corresponding to each of the hyperparameters according to the influence amplitude value corresponding to the current hyperparameter to be adjusted.

[0227] Furthermore, before determining the adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the timing characteristic category information of the network indicators, it also includes: determining the timing characteristic category information of the network indicators according to the timing characteristic parameter information of the network indicators; wherein the timing characteristic parameter information includes: at least one of indicator direction information, timing characteristic information, value type information and value order of magnitude information.

[0228] Information on indicator direction may include positive, negative or neutral, etc.; time series feature information may include daily cycle × trend, daily cycle + weekly cycle, daily cycle + trend, single value + abnormal point or stable sequence, etc.; value type information may include absolute value or ratio value, etc.; value magnitude may be expressed as ratio, single digit level, tens digit level, hundreds digit level, thousands digit level or tens of thousands digit level, etc.

[0229] Furthermore, after updating the current value of the current hyperparameter to be adjusted according to the initial target value, it also includes: determining a group of first sampling points according to the initial target value, the adjustment step and the number of first sampling points; the first sampling points are located on both sides of the initial target value; obtaining the effect parameter values ​​corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value; and determining the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value.

[0230] Among them, determining the final target value of the current hyperparameter to be adjusted based on the first optimal effect parameter value and the first worst effect parameter value includes: when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is less than or equal to a first threshold, increasing the adjustment step according to a preset amplitude, and returning to execute the step of determining a group of first sampling points based on the initial target value and the adjustment step according to the increased adjustment step; until the adjustment step reaches the step upper limit, obtaining the target effect parameter value from the respective effect parameter values ​​corresponding to the respective first sampling points, and taking the value of the first sampling point corresponding to the target effect parameter value as the final target value of the current hyperparameter to be adjusted.

[0231] The target effect parameter value can also be understood as the optimal value.

[0232] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted based on the first optimal effect parameter value and the first worst effect parameter value includes: when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is greater than a first threshold, determining the search step, the search direction and the search endpoint; wherein the search step is equal to the adjustment step, and the search endpoint is the first sampling point at the end of the search direction; determining the final target value of the current hyperparameter to be adjusted based on the search step, the search direction and the search endpoint.

[0233] Wherein, determining the search direction includes: determining the search direction according to the initial target value and the first sampling point corresponding to the optimal effect parameter value.

[0234] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted according to the search step, the search direction, and the search endpoint includes: determining a group of second sampling points according to the search step, the search direction, the search endpoint, and the number of second sampling points; obtaining the effect parameter value corresponding to each of the second sampling points to obtain a second optimal effect parameter value and a second worst effect parameter value; and determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value.

[0235] Specifically, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to the third threshold, re-determining the search step, search direction and search endpoint; according to the re-determined search step, search direction and search endpoint, returning to execute the determination of the final target value of the current hyperparameter to be adjusted according to the search step, search direction and search endpoint; wherein the re-determined search direction is consistent with the previous search direction; the search step is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; the search endpoint is the second sampling point located at the end of the search direction; the second threshold is less than the first threshold.

[0236] The “sampling point of the previous batch” can be the first sampling point or the second sampling point; the “difference” can be a positive value (that is, an increase), such as in the case of accuracy; or a negative value (that is, a decrease), such as in the case of mean square error.

[0237] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted based on the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to a second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a third threshold, determining that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted; wherein the second threshold is less than the first threshold.

[0238] The “sampling point of the previous batch” can be the first sampling point or the second sampling point; the “difference” can be a positive value (that is, an increase), such as in the case of accuracy; or a negative value (that is, a decrease), such as in the case of mean square error.

[0239] In an embodiment of the present invention, the determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value may include: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to a second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a third threshold, obtaining a target effect parameter value from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and taking the value of the sampling point corresponding to the target effect parameter value as the final target value of the current hyperparameter to be adjusted.

[0240] The "previous batch sampling point" can be the first sampling point or the second sampling point; the "difference" can be a positive value (that is, an increase), such as for the accuracy; or a negative value (that is, a decrease), such as for the mean square error. The target effect parameter value can also be understood as the optimal value.

[0241] Wherein, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is greater than a second threshold, redetermining the search direction; determining the final target value of the current hyperparameter to be adjusted according to the redetermined search direction; wherein the redetermined search direction is the direction from the second sampling point corresponding to the second worst effect parameter value to the second sampling point corresponding to the second optimal effect parameter value; and the second threshold is less than the first threshold.

[0242] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: when the re-determined search direction is consistent with the previous search direction, or the re-determined search direction is not consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to a fourth threshold, re-determining the search step and the search endpoint; according to the re-determined search step, search direction and search endpoint, returning to execute the determination of the final target value of the current hyperparameter to be adjusted according to the search step, search direction and search endpoint; wherein the search step is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; the search endpoint is the second sampling point located at the end of the search direction; and the second threshold is less than the first threshold.

[0243] The "previous batch sampling point" can be the first sampling point or the second sampling point; the "difference" can be a positive value (that is, an increase), such as for the accuracy; or a negative value (that is, a decrease), such as for the mean square error. The target effect parameter value can also be understood as the optimal value.

[0244] Among them, determining the final target value of the current hyperparameter to be adjusted based on the re-determined search direction includes: when the re-determined search direction is inconsistent with the direction of the previous search, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, determining that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted.

[0245] The “sampling point of the previous batch” can be the first sampling point or the second sampling point; the “difference” can be a positive value (that is, an increase), such as in the case of accuracy; or a negative value (that is, a decrease), such as in the case of mean square error.

[0246] In an embodiment of the present invention, the determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction may include: when the re-determined search direction is consistent with the direction of the previous search, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, obtaining a target effect parameter value from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and taking the value of the sampling point corresponding to the target effect parameter value as the final target value of the current hyperparameter to be adjusted.

[0247] The "previous batch sampling point" can be the first sampling point or the second sampling point; the "difference" can be a positive value (that is, an increase), such as for the accuracy; or a negative value (that is, a decrease), such as for the mean square error. The target effect parameter value can also be understood as the optimal value.

[0248] The information processing method provided by the embodiment of the present invention is further described below.

[0249] In response to the above technical problems, an embodiment of the present invention provides an information processing method, which can be specifically implemented as an automatic parameter adjustment method based on the Prophet algorithm, which can realize automatic parameter adjustment for different timing indicators in the network, obtain better parameter configuration, and thus improve the performance of the algorithm.

[0250] The solution provided in the embodiment of the present invention includes two parts: coarse parameter adjustment and fine parameter adjustment. In actual use, only one part may be used, or both parts may be used in combination. Figure 2 As shown, this program involves:

[0251] 1. Rough adjustment process:

[0252] 1. Classification of time series features;

[0253] Specifically, the temporal characteristics of network indicators are classified into four categories: indicator direction (such as positive, negative, neutral), temporal characteristics (which may include the following categories), value type (such as absolute value, ratio value), and value magnitude (such as ratio, one / ten / hundred / thousand / ten thousand...);

[0254] Time series feature categories: daily cycle × trend, daily cycle + weekly cycle, daily cycle + trend, single value + abnormal point, stable sequence;

[0255] 2. Parameter adjustment template selection;

[0256] Specifically, a parameter adjustment template is selected according to the classification result. The template content includes the adjustment order of the hyperparameters and the set of sampling scale values ​​(i.e., the candidate values) of each hyperparameter.

[0257] For the parameter adjustment order of different timing categories, please refer to Table 1 below:

[0258] Table 1

[0259]

[0260] The horizontal direction (a row) of the table represents the adjustment number of the same hyperparameter under different time series categories, and the vertical direction (a column) of the table represents the adjustment number of each hyperparameter involved in the same time series category.

[0261] Time series category 1 is single value + irregulars, time series category 2 is stationary, time series category 3 is daily seasonal, time series category 4 is nonmonotrend + daily seasonal, time series category 5 is daily + weekly seasonal, time series category 6 is step + daily seasonal, time series category 7 is daily seasonal × trend, time series category 8 is monotrend + daily seasonal. seasonal (single trend + daily cycle); parameter 1 is growth, parameter 2 is mcmc_samples (Markov Chain Monte Carlo method samples), parameter 3 is changepoint_range (change point range), parameter 4 is n_changepoints (number of change points), parameter 5 is weekly_seasonality (weekly cycle), parameter 6 is daily_seasonality (daily cycle), parameter 7 is seasonality_mode (cycle mode), parameter 8 is seasonality_prior_scale (seasonal prior probability), parameter 9 is changepoint_prior_scale (change point prior probability).

[0262] For the sampling scale value set of each hyperparameter, see Table 2 below:

[0263] Table 2

[0264] Paras scale value growth linear, logistic mcmc_samples 0 changepoint_range 0.8，0.9，1.0 n_changepoints 5，25，50，250 weekly_seasonality 3，30，300 daily_seasonality 4，40，400 seasonality_mode additive, multiplicative seasonality_prior_scale 0.01，0.5，5.0，10.0，50.0，250.0，2500.0 changepoint_prior_scale 0.01，0.05，0.5，5.0，50.0，250.0

[0265] 3. Adjust parameters and sort;

[0266] Specifically, select the hyperparameters to be adjusted in order according to the template sequence, fix other hyperparameters to the default values ​​or the adjusted optimal values, evaluate the algorithm effect (the evaluation parameters may include prediction accuracy and / or root mean square error RMSE), and determine the optimal value of the hyperparameter. Complete the evaluation of all hyperparameters in this way to obtain a set of optimal hyperparameter configurations.

[0267] 4. Effect evaluation (rmse, cov (prediction accuracy));

[0268] Specifically, calculate the impact of each hyperparameter on the algorithm effect (there are many ways to calculate the impact, such as abs (algorithm effect under the optimal value - algorithm effect under the default value)), and rearrange the parameter adjustment order from large to small according to the impact.

[0269] Among them, the algorithm effect after each hyperparameter is adjusted can be compared with the algorithm effect before the hyperparameter is adjusted.

[0270] 5. Coarse tuning results and hyper-parameter tuning sequence;

[0271] Specifically, the above operation 3 may be repeated at least once to obtain a rough adjustment result, including a parameter adjustment sequence and a preferred rough adjustment value of each parameter.

[0272] 2. Fine-tuning process:

[0273] 1. Select hyperparameters;

[0274] 2. Set the starting point, step size, search direction, search range, etc.;

[0275] Specifically, (based on experience) different step sizes are configured or calculated for different scale values ​​of different parameters, the number of calculation points for each search (i.e., the number of sampling points) is set, a delta threshold (thres, i.e., the first threshold mentioned above) for determining whether a local depression is found and a delta threshold (ultra-thres, i.e., the second threshold mentioned above, which is less than thres) for determining whether a local optimal depression is found are set;

[0276] Subsequently, taking the optimal value of the hyper-parameter after rough adjustment as the starting point, according to the parameter adjustment order determined after rough adjustment, each hyper-parameter is automatically searched according to the fine-tuning process:

[0277] 3. Effect evaluation (rmse, cov); determine whether a local depression is found;

[0278] If not, increase the step size, adjust the search direction, etc., and return the execution effect evaluation (rmse, cov);

[0279] If yes, enter the gradient estimation to determine whether the stop condition is met; if yes, get the fine-tuning result and return to select the hyperparameter; if no, adjust the starting point, step size, search direction, search range, etc., and return to perform gradient estimation.

[0280] Specifically, it involves the following:

[0281] (1) Global search:

[0282] 1) According to the step size corresponding to the scale, search both sides at the same time (corresponding to the first sampling point being located on both sides of the initial target value), and calculate the corresponding algorithm performance (i.e., the effect parameter value) at each sampling point. Suppose the best performance of the batch of sampling points is vmax, and the worst performance is vmin (of course, the best performance can also be vmin, and the worst performance can be vmax, which is not limited here), then:

[0283] a) If abs(vmax - vmin) ≤ thres, increase the step size by a preset amplitude (i.e., the above-adjusted step size), and perform operation 1) again until the maximum step size is reached and exited (i.e., meeting the process stop condition 3), and select the parameter value corresponding to the best performance as the fine-tuning result;

[0284] b) If abs(vmax - vmin) > thres, it means that a local depression is found, and the search direction (decrease or increase), step size (the current step size, i.e., the above search step size), and search starting point (the end point of the current batch in the search direction, i.e., the above search end point) are re-determined;

[0285] (2) Local search (based on global search):

[0286] Based on the number of sampling points (i.e., the above second sampling point quantity) and the search direction, step size, and search starting point obtained from the global search, each sampling point (i.e., the above second sampling point) is obtained; the algorithm performance on each batch of sampling points is evaluated, the performance delta = abs(vmax - vmin) of this batch is calculated, and a judgment is made (i.e., the above gradient estimation), where the performance optimal value vmax (for the performance optimal value, it can be that the larger the better, such as accuracy; or the smaller the better, such as mean square error):

[0287] 1) If delta < ultra - thres: It is judged that this batch is in the performance saddle point area (flat area), and at this time, the judgment is as follows:

[0288] a) If vmax decreases by more than the threshold (i.e., the above third threshold) compared with vmax of the previous batch, the search direction is the same as that of the previous batch, the starting point is set as the end point corresponding to the search direction of this batch, the step size is updated, a new batch of sampling points is obtained, and the process returns to perform the evaluation of the algorithm performance on each batch of sampling points, calculate the performance delta of this batch, and make a judgment; specifically, the step size = abs(paramax of the sampling point corresponding to the performance optimal value - paramin of the sampling point corresponding to the performance worst value);

[0289] b) Otherwise, it is determined that a local better point is found, and the program exits (i.e., the stop condition 1 is met); the parameter value corresponding to the best performance is selected as the fine-tuning result (i.e., when the absolute value of the difference between the second best effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second best effect parameter value and the best effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, the value of the sampling point corresponding to the second best effect parameter value is determined to be the final target value of the hyperparameter to be adjusted; or, when the absolute value of the difference between the second best effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second best effect parameter value and the best effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the hyperparameter to be adjusted);

[0290] 2) If delta>ultra-thres: the batch is judged to be in the gradient effective area, and the search direction of the batch is determined to be from vmin to vmax (that is, the search direction is re-determined when the absolute value of the difference between the second best effect parameter value and the second worst effect parameter value is greater than the second threshold value). At this time, it is judged that:

[0291] a) If the search direction of this batch is the same as that of the previous batch, keep the search direction unchanged, and update the starting point and step size (corresponding to the above-mentioned redetermined search direction is consistent with the previous search direction, redetermine the search step size and search endpoint), obtain a new batch of sampling points, return to execute the algorithm performance evaluation on each batch of sampling points, calculate the performance delta on the batch, and make a judgment; specifically, step size = abs (sampling point paramax corresponding to the optimal performance value - sampling point paramin corresponding to the worst performance value);

[0292] b) If the search direction of this batch is different from that of the previous batch, but the decrease of vmax exceeds the threshold (i.e., the fourth threshold mentioned above), the overall trend is judged to be decreasing, and the search continues along the search direction of the previous batch: the search direction is set to be the same as that of the previous batch, and the starting point is the endpoint corresponding to the search direction of this batch, and the step size is updated (i.e., when the re-determined search direction is not consistent with the direction of the previous search, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to the fourth threshold, the search step size and the search endpoint are re-determined), a new batch of sampling points are obtained, and the algorithm performance on each batch of sampling points is evaluated, the performance delta on the batch is calculated, and a judgment is made; specifically, step size = abs (sampling point paramax corresponding to the optimal performance value - sampling point paramin corresponding to the worst performance value);

[0293] c) Otherwise, it is determined that a local better point is found, and the program exits (i.e., the stop condition 2 is met); the parameter value corresponding to the best performance is selected as the fine-tuning result (i.e., when the re-determined search direction is consistent with the direction of the previous search, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the fourth threshold, the value of the sampling point corresponding to the second optimal effect parameter value is determined to be the final target value of the current hyperparameter to be adjusted; or, when the re-determined search direction is consistent with the direction of the previous search, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the fourth threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted).

[0294] As can be seen from the above, the solution provided by the embodiment of the present invention involves:

[0295] (1): The above parameter coarse adjustment and parameter fine adjustment process mainly involves:

[0296] Classify network indicators, select parameter adjustment templates according to the classification results, sort the hyperparameter adjustment sequence multiple times according to expert experience or algorithm operation results, and make rough adjustments based on the last sorting result to generate a hyperparameter adjustment sequence and a set of hyperparameter values;

[0297] Fine tuning is performed based on the coarse tuning results, including global search and local search processes. In global search, a variable step size is used to find the local depression area. In local search, the number of sampling points in each batch is configured. According to the algorithm performance on the sampling points, the <search direction, search step size, search starting point> triples of the next batch are determined, and finally the fine tuning results are obtained.

[0298] (2): Classify network indicators:

[0299] The classification basis includes the indicator direction (including positive, negative, neutral), time series characteristics, and may also include value type (absolute value, ratio value), value magnitude (ratio, pieces / tens / hundreds / thousands / ten thousand...), etc.

[0300] Among them, time series features include daily cycle × trend, daily cycle + weekly cycle, daily cycle + trend, and can also include single value + abnormal point, stable sequence, etc.

[0301] (3): Select the parameter adjustment template according to the classification results:

[0302] The template content includes the adjustment order of hyperparameters and the sampling value set of each hyperparameter;

[0303] (4): Rearrange the order of adjusting parameters according to the influence of each hyperparameter on the algorithm effect from large to small;

[0304] (5): Different step sizes can be set for different scale values ​​of different parameters. During global search, the step size is adjusted based on the judgment threshold.

[0305] In summary, the solution provided by the embodiment of the present invention can realize the automatic parameter adjustment of network timing indicators, and the parameter adjustment effect is significant. For timing with obvious daily cycles, weekly cycles or trends, the performance difference before and after prophet hyperparameter optimization is large, for example: for C2000030: the number of ATS registered users, coarse adjustment can improve the performance by about 56%, and fine adjustment can further improve the performance by about 10%; for C2000136: the number of successful SBC re-registrations, coarse adjustment can improve the performance by about 40%, and fine adjustment can further improve the performance by about 2.5%.

[0306] Embodiment 2

[0307] The information processing method provided by the embodiment of the present invention is as follows: Figure 3 As shown, including:

[0308] Step 31: determining a set of first sampling points according to the initial target value of the hyperparameter to be adjusted, the adjustment step size, and the number of first sampling points; the first sampling points are located on both sides of the initial target value;

[0309] Step 32: Obtain the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value;

[0310] Step 33: Determine the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value.

[0311] Among them, the determination of the "current hyperparameter to be adjusted" can be specifically: determining the adjustment order of the hyperparameters for the network indicators based on the timing feature category information of the network indicators; and determining the current hyperparameter to be adjusted from each of the hyperparameters based on the adjustment order.

[0312] The “initial target value” may be a default value or a value after preliminary adjustment (rough adjustment), and is not limited here.

[0313] The information processing method provided by the embodiment of the present invention determines a group of first sampling points according to the initial target value, adjustment step size and number of first sampling points of the current hyperparameter to be adjusted; the first sampling points are located on both sides of the initial target value; the effect parameter values ​​corresponding to each of the first sampling points are obtained to obtain a first optimal effect parameter value and a first worst effect parameter value; the final target value of the current hyperparameter to be adjusted is determined according to the first optimal effect parameter value and the first worst effect parameter value; it can realize automatic parameter adjustment for different timing indicators in the network, avoiding heavy workload and waste of manpower; and it can obtain better parameter configuration, thereby improving the performance of the algorithm and enhancing the adjustment effect; it also realizes automatic adjustment of hyperparameters (it can also realize automatic parameter adjustment based on the Prophet algorithm); it well solves the problems of heavy workload, waste of manpower, inability to realize automation and poor adjustment effect in the information processing scheme for hyperparameter adjustment of the Prophet algorithm for network indicators in the prior art.

[0314] It is explained here that the solution of the second embodiment of the present invention can refer to the relevant content of the solution of the first embodiment of the present invention, and is not limited here.

[0315] The embodiment of the present invention also provides an information processing device, such as Figure 4 As shown, including:

[0316] A first determination module 41 is used to determine the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the time series feature category information of the network indicators;

[0317] A second determination module 42 is used to determine the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order;

[0318] A first processing module 43 is used to obtain an initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters;

[0319] A first updating module 44 is used to update the current value of the hyperparameter to be adjusted according to the initial target value;

[0320] The other hyperparameters are the hyperparameters of the network indicators except the hyperparameters to be adjusted currently.

[0321] The information processing device provided in the embodiment of the present invention determines the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the timing feature category information of the network indicators; determines the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order; obtains the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current value of other hyperparameters; updates the current value of the current hyperparameter to be adjusted according to the initial target value; wherein the other hyperparameters are the parameters of the hyperparameters of the network indicators other than the current hyperparameter to be adjusted; it can realize automatic parameter adjustment for different timing indicators in the network, avoiding heavy workload and waste of manpower; and it can obtain better parameter configuration, thereby improving the performance of the algorithm and enhancing the adjustment effect; it also realizes automatic adjustment of hyperparameters (it can also realize automatic parameter adjustment based on the Prophet algorithm); it well solves the problems of heavy workload, waste of manpower, inability to realize automation and poor adjustment effect in the information processing scheme for hyperparameter adjustment of the Prophet algorithm for network indicators in the prior art.

[0322] Among them, the first processing module includes: a first processing sub-module, which is used to obtain the effect parameter value of the network indicator corresponding to each candidate value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current value of other hyperparameters; a first determination sub-module, which is used to determine the initial target value of the current hyperparameter to be adjusted from each candidate value of the current hyperparameter to be adjusted according to the effect parameter value.

[0323] Furthermore, the information processing device also includes: a first acquisition module, used to obtain the influence amplitude value corresponding to each of the hyperparameters after updating the current value of the current hyperparameter to be adjusted according to the initial target value; a second updating module, used to update the parameter adjustment sequence according to the influence amplitude value; a second processing module, used to return to execute the step of determining the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment sequence after the update; obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; and updating the current value of the current hyperparameter to be adjusted according to the initial target value.

[0324] Among them, the first acquisition module includes: a second processing sub-module, which is used to obtain the influence amplitude value corresponding to the current hyperparameter to be adjusted according to the effect parameter value corresponding to the current value of the hyperparameter to be adjusted before the current value is updated and the effect parameter value corresponding to the current value of the hyperparameter to be adjusted after the current value is updated; a third processing sub-module, which is used to obtain the influence amplitude value corresponding to each of the hyperparameters according to the influence amplitude value corresponding to the current hyperparameter to be adjusted.

[0325] Furthermore, the information processing device also includes: a third determination module, which is used to determine the timing characteristic category information of the network indicator according to the timing characteristic parameter information of the network indicator before determining the adjustment order of the hyperparameters of the network indicator and the candidate values ​​of each of the hyperparameters according to the timing characteristic category information of the network indicator; wherein the timing characteristic parameter information includes: at least one of indicator direction information, timing characteristic information, value type information and value order of magnitude information.

[0326] In an embodiment of the present invention, the information processing device further includes: a fourth determination module, which is used to determine a group of first sampling points according to the initial target value, the adjustment step size and the number of first sampling points after updating the current value of the current hyperparameter to be adjusted according to the initial target value; the first sampling points are located on both sides of the initial target value; a second acquisition module, which is used to obtain the effect parameter values ​​corresponding to each of the first sampling points to obtain the first optimal effect parameter value and the first worst effect parameter value; a fifth determination module, which is used to determine the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value.

[0327] Among them, the fifth determination module includes: a fourth processing sub-module, which is used to increase the adjustment step according to a preset amplitude when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is less than or equal to the first threshold, and return to execute the step of determining a group of first sampling points according to the initial target value and the adjustment step according to the increased adjustment step; until the adjustment step reaches the step upper limit, the target effect parameter value is obtained from the effect parameter values ​​corresponding to the obtained first sampling points, and the value of the first sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

[0328] Specifically, the fifth determination module includes: a second determination submodule, used to determine the search step, search direction and search endpoint when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is greater than a first threshold; wherein the search step is equal to the adjustment step, and the search endpoint is the first sampling point at the end of the search direction; and a third determination submodule, used to determine the final target value of the current hyperparameter to be adjusted based on the search step, search direction and search endpoint.

[0329] Wherein, determining the search direction includes: determining the search direction according to the initial target value and the first sampling point corresponding to the optimal effect parameter value.

[0330] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted according to the search step, the search direction, and the search endpoint includes: determining a group of second sampling points according to the search step, the search direction, the search endpoint, and the number of second sampling points; obtaining the effect parameter value corresponding to each of the second sampling points to obtain a second optimal effect parameter value and a second worst effect parameter value; and determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value.

[0331] Wherein, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to the third threshold, re-determining the search step, search direction and search endpoint; according to the re-determined search step, search direction and search endpoint, returning to execute the determination of the final target value of the current hyperparameter to be adjusted according to the search step, search direction and search endpoint; wherein the re-determined search direction is consistent with the last search direction; the search step is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; the search endpoint is the second sampling point located at the end of the search direction; the second threshold is less than the first threshold.

[0332] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted based on the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to a second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a third threshold, determining that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted; wherein the second threshold is less than the first threshold.

[0333] Among them, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to a second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a third threshold, obtaining a target effect parameter value from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and taking the value of the sampling point corresponding to the target effect parameter value as the final target value of the current hyperparameter to be adjusted.

[0334] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted based on the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is greater than a second threshold, redetermining the search direction; determining the final target value of the current hyperparameter to be adjusted based on the redetermined search direction; wherein the redetermined search direction is the direction from the second sampling point corresponding to the second worst effect parameter value to the second sampling point corresponding to the second optimal effect parameter value; and the second threshold is less than the first threshold.

[0335] Wherein, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: when the re-determined search direction is consistent with the previous search direction, or the re-determined search direction is not consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to a fourth threshold, re-determining the search step and the search endpoint; according to the re-determined search step, search direction and search endpoint, returning to execute the determination of the final target value of the current hyperparameter to be adjusted according to the search step, search direction and search endpoint; wherein the search step is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; the search endpoint is the second sampling point located at the end of the search direction; the second threshold is less than the first threshold.

[0336] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted based on the re-determined search direction includes: when the re-determined search direction is inconsistent with the direction of the previous search, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, determining that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted.

[0337] Among them, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: when the re-determined search direction is consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, obtaining a target effect parameter value from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and taking the value of the sampling point corresponding to the target effect parameter value as the final target value of the current hyperparameter to be adjusted.

[0338] Among them, the implementation embodiments of the information processing method of the above-mentioned embodiment 1 are all applicable to the embodiments of the information processing device and can also achieve the same technical effects.

[0339] The embodiment of the present invention also provides an information processing device, such as Figure 5 As shown, including:

[0340] A sixth determination module 51 is used to determine a set of first sampling points according to the initial target value of the hyperparameter to be adjusted, the adjustment step size, and the number of first sampling points; the first sampling points are located on both sides of the initial target value;

[0341] A third acquisition module 52 is used to acquire the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value;

[0342] The seventh determination module 53 is used to determine the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value.

[0343] The information processing device provided by the embodiment of the present invention determines a group of first sampling points according to the initial target value, adjustment step size and number of first sampling points of the current hyperparameter to be adjusted; the first sampling points are located on both sides of the initial target value; the effect parameter values ​​corresponding to each of the first sampling points are obtained to obtain the first optimal effect parameter value and the first worst effect parameter value; the final target value of the current hyperparameter to be adjusted is determined according to the first optimal effect parameter value and the first worst effect parameter value; it can realize automatic parameter adjustment for different timing indicators in the network, avoiding heavy workload and waste of manpower; and it can obtain better parameter configuration, thereby improving the performance of the algorithm and enhancing the adjustment effect; it also realizes automatic adjustment of hyperparameters (it can also realize automatic parameter adjustment based on the Prophet algorithm); it well solves the problems of heavy workload, waste of manpower, inability to realize automation and poor adjustment effect in the information processing scheme for hyperparameter adjustment of the Prophet algorithm for network indicators in the prior art.

[0344] Among them, the implementation embodiments of the information processing method of the above-mentioned embodiment 2 are all applicable to the embodiments of the information processing device and can also achieve the same technical effects.

[0345] The embodiment of the present invention also provides an information processing device, such as Figure 6 As shown, it includes: a processor 61;

[0346] The processor 61 is used to determine the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the time series feature category information of the network indicators;

[0347] According to the parameter adjustment order, determine the current hyperparameter to be adjusted from each of the hyperparameters;

[0348] Obtaining an initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters;

[0349] Updating the current value of the current hyperparameter to be adjusted according to the initial target value;

[0350] The other hyperparameters are the hyperparameters of the network indicators except the hyperparameters to be adjusted currently.

[0351] In the embodiment of the present invention, the information processing device may further include a transceiver for communicating with the processor, which is not limited here.

[0352] The information processing device provided by the embodiment of the present invention determines the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the timing feature category information of the network indicators; determines the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order; obtains the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current value of other hyperparameters; updates the current value of the current hyperparameter to be adjusted according to the initial target value; wherein the other hyperparameters are the parameters of the hyperparameters of the network indicators other than the current hyperparameter to be adjusted; it can realize automatic parameter adjustment for different timing indicators in the network, avoiding heavy workload and waste of manpower; and it can obtain better parameter configuration, thereby improving the performance of the algorithm and enhancing the adjustment effect; it also realizes automatic adjustment of hyperparameters (it can also realize automatic parameter adjustment based on the Prophet algorithm); it well solves the problems of heavy workload, waste of manpower, inability to realize automation and poor adjustment effect in the information processing scheme for hyperparameter adjustment of the Prophet algorithm for network indicators in the prior art.

[0353] Among them, obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters includes: obtaining the effect parameter value of the network indicator corresponding to each candidate value of the current hyperparameter to be adjusted according to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; and determining the initial target value of the current hyperparameter to be adjusted from each candidate value of the current hyperparameter to be adjusted according to the effect parameter value.

[0354] Furthermore, the processor is also used to: after updating the current value of the current hyperparameter to be adjusted according to the initial target value, obtain the influence amplitude value corresponding to each of the hyperparameters; update the parameter adjustment order according to the influence amplitude value; according to the updated parameter adjustment order, return to execute the process of determining the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order; obtain the initial target value of the current hyperparameter to be adjusted according to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; and update the current value of the current hyperparameter to be adjusted according to the initial target value.

[0355] Among them, the obtaining of the influence amplitude value corresponding to each of the hyperparameters includes: obtaining the influence amplitude value corresponding to the current hyperparameter to be adjusted according to the effect parameter value corresponding to the current value of the hyperparameter to be adjusted before being updated and the effect parameter value corresponding to the current value of the hyperparameter to be adjusted after being updated; obtaining the influence amplitude value corresponding to each of the hyperparameters according to the influence amplitude value corresponding to the current hyperparameter to be adjusted.

[0356] Furthermore, the processor is also used to: determine the timing characteristic category information of the network indicator according to the timing characteristic parameter information of the network indicator before determining the adjustment order of the hyperparameters of the network indicator and the candidate values ​​of each of the hyperparameters according to the timing characteristic category information of the network indicator; wherein the timing characteristic parameter information includes: at least one of indicator direction information, timing characteristic information, value type information and value order of magnitude information.

[0357] In an embodiment of the present invention, the processor is also used to: after updating the current value of the current hyperparameter to be adjusted according to the initial target value, determine a group of first sampling points according to the initial target value, the adjustment step size and the number of first sampling points; the first sampling points are located on both sides of the initial target value; obtain the effect parameter values ​​corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value; determine the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value.

[0358] Among them, determining the final target value of the current hyperparameter to be adjusted based on the first optimal effect parameter value and the first worst effect parameter value includes: when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is less than or equal to a first threshold, increasing the adjustment step according to a preset amplitude, and returning to execute the step of determining a group of first sampling points based on the initial target value and the adjustment step according to the increased adjustment step; until the adjustment step reaches the step upper limit, obtaining the target effect parameter value from the respective effect parameter values ​​corresponding to the respective first sampling points, and taking the value of the first sampling point corresponding to the target effect parameter value as the final target value of the current hyperparameter to be adjusted.

[0359] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted based on the first optimal effect parameter value and the first worst effect parameter value includes: when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is greater than a first threshold, determining the search step, the search direction and the search endpoint; wherein the search step is equal to the adjustment step, and the search endpoint is the first sampling point at the end of the search direction; determining the final target value of the current hyperparameter to be adjusted based on the search step, the search direction and the search endpoint.

[0360] Wherein, determining the search direction includes: determining the search direction according to the initial target value and the first sampling point corresponding to the optimal effect parameter value.

[0361] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted according to the search step, the search direction, and the search endpoint includes: determining a group of second sampling points according to the search step, the search direction, the search endpoint, and the number of second sampling points; obtaining the effect parameter value corresponding to each of the second sampling points to obtain a second optimal effect parameter value and a second worst effect parameter value; and determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value.

[0362] Wherein, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to the third threshold, re-determining the search step, search direction and search endpoint; according to the re-determined search step, search direction and search endpoint, returning to execute the determination of the final target value of the current hyperparameter to be adjusted according to the search step, search direction and search endpoint; wherein the re-determined search direction is consistent with the last search direction; the search step is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; the search endpoint is the second sampling point located at the end of the search direction; the second threshold is less than the first threshold.

[0363] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted based on the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to a second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a third threshold, determining that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted; wherein the second threshold is less than the first threshold.

[0364] Among them, determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to a second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a third threshold, obtaining a target effect parameter value from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and taking the value of the sampling point corresponding to the target effect parameter value as the final target value of the current hyperparameter to be adjusted.

[0365] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted based on the second optimal effect parameter value and the second worst effect parameter value includes: when the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is greater than a second threshold, redetermining the search direction; determining the final target value of the current hyperparameter to be adjusted based on the redetermined search direction; wherein the redetermined search direction is the direction from the second sampling point corresponding to the second worst effect parameter value to the second sampling point corresponding to the second optimal effect parameter value; and the second threshold is less than the first threshold.

[0366] Wherein, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: when the re-determined search direction is consistent with the previous search direction, or the re-determined search direction is not consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to a fourth threshold, re-determining the search step and the search endpoint; according to the re-determined search step, search direction and search endpoint, returning to execute the determination of the final target value of the current hyperparameter to be adjusted according to the search step, search direction and search endpoint; wherein the search step is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; the search endpoint is the second sampling point located at the end of the search direction; the second threshold is less than the first threshold.

[0367] In an embodiment of the present invention, determining the final target value of the current hyperparameter to be adjusted based on the re-determined search direction includes: when the re-determined search direction is inconsistent with the direction of the previous search, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, determining that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted.

[0368] Among them, determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: when the re-determined search direction is consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, obtaining a target effect parameter value from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and taking the value of the sampling point corresponding to the target effect parameter value as the final target value of the current hyperparameter to be adjusted.

[0369] Among them, the implementation embodiments of the information processing method of the above-mentioned embodiment 1 are all applicable to the embodiments of the information processing device and can also achieve the same technical effects.

[0370] The embodiment of the present invention also provides an information processing device, such as Figure 7 As shown, it includes: a processor 71;

[0371] The processor 71 is used to determine a set of first sampling points according to the initial target value of the hyperparameter to be adjusted, the adjustment step size and the number of first sampling points; the first sampling points are located on both sides of the initial target value;

[0372] Obtaining the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value;

[0373] According to the first optimal effect parameter value and the first worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

[0374] In the embodiment of the present invention, the information processing device may further include a transceiver for communicating with the processor, which is not limited here.

[0375] The information processing device provided by the embodiment of the present invention determines a group of first sampling points according to the initial target value, adjustment step size and number of first sampling points of the current hyperparameter to be adjusted; the first sampling points are located on both sides of the initial target value; the effect parameter values ​​corresponding to each of the first sampling points are obtained to obtain the first optimal effect parameter value and the first worst effect parameter value; the final target value of the current hyperparameter to be adjusted is determined according to the first optimal effect parameter value and the first worst effect parameter value; it can realize automatic parameter adjustment for different timing indicators in the network to avoid heavy workload and waste of manpower; and it can obtain better parameter configuration, thereby improving the performance of the algorithm and enhancing the adjustment effect; it also realizes automatic adjustment of hyperparameters (it can also realize automatic parameter adjustment based on the Prophet algorithm); it well solves the problems of heavy workload, waste of manpower, inability to realize automation and poor adjustment effect in the information processing scheme for hyperparameter adjustment of the Prophet algorithm for network indicators in the prior art.

[0376] Among them, the implementation embodiments of the information processing method of the above-mentioned embodiment 2 are all applicable to the embodiments of the information processing device and can also achieve the same technical effects.

[0377] The embodiment of the present invention further provides an information processing device, comprising a memory, a processor, and a program stored in the memory and executable on the processor; when the processor executes the program, the information processing method described in the first embodiment is implemented; or,

[0378] When the processor executes the program, the information processing method described in the above-mentioned embodiment 2 is implemented.

[0379] Among them, the implementation embodiments of the information processing method of the above-mentioned embodiment 1 or embodiment 2 are all applicable to the embodiments of the information processing device and can also achieve the corresponding same technical effects.

[0380] The embodiment of the present invention further provides a readable storage medium on which a program is stored. When the program is executed by a processor, the steps of the information processing method described in the first embodiment are implemented; or,

[0381] When the program is executed by a processor, the steps of the information processing method described in the above-mentioned embodiment 2 are implemented.

[0382] Among them, the implementation embodiments of the information processing method of the above-mentioned embodiment 1 or embodiment 2 are all applicable to the embodiments of the readable storage medium, and can also achieve the corresponding same technical effects.

[0383] It should be noted that many functional components described in this specification are referred to as modules / sub-modules in order to more particularly emphasize the independence of their implementation methods.

[0384] In the embodiment of the present invention, module / submodule can be implemented with software so that it can be performed by various types of processors. For example, an executable code module of an identification can include one or more physical or logical blocks of computer instructions, for example, it can be constructed as an object, process or function. Nevertheless, the executable code of the identified module does not need to be physically located together, but can include different instructions stored in different positions, and when these instructions are logically combined together, it constitutes a module and realizes the specified purpose of this module.

[0385] In fact, executable code module can be a single instruction or many instructions, and can even be distributed on a plurality of different code segments, distributed among different programs, and distributed across a plurality of memory devices. Similarly, operating data can be identified in the module, and can be implemented and organized in the data structure of any appropriate type according to any appropriate form. The operating data can be collected as a single data set, or can be distributed in different locations (including on different storage devices), and can only be present on a system or network as an electronic signal at least in part.

[0386] When a module can be implemented by software, considering the level of existing hardware technology, a person skilled in the art can build a corresponding hardware circuit to implement the corresponding function of the module that can be implemented by software without considering the cost. The hardware circuit includes a conventional very large scale integration (VLSI) circuit or gate array and existing semiconductors such as logic chips, transistors, or other discrete components. The module can also be implemented by a programmable hardware device, such as a field programmable gate array, a programmable array logic, a programmable logic device, etc.

[0387] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary personnel in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An information processing method, characterized in that: include: Determine, according to the time series feature category information of the network indicator, the parameter adjustment order of the hyperparameters of the network indicator and the candidate values ​​of each of the hyperparameters; According to the parameter adjustment order, determine the current hyperparameter to be adjusted from each of the hyperparameters; Obtaining an initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; Updating the current value of the current hyperparameter to be adjusted according to the initial target value; The other hyperparameters are the hyperparameters of the network indicators except the hyperparameters to be adjusted currently. Before determining the adjustment order of the hyperparameters of the network indicators and the candidate values ​​of the hyperparameters according to the time series feature category information of the network indicators, the method further includes: Determine the timing characteristic category information of the network indicator according to the timing characteristic parameter information of the network indicator; wherein the timing characteristic parameter information includes: at least one of indicator direction information, timing characteristic information, value type information and value magnitude information; The step of determining the adjustment order of the hyperparameters of the network indicators and the candidate values ​​of the hyperparameters according to the time series feature category information of the network indicators includes: According to the time series feature category information of the network indicator and the pre-stored corresponding relationship information, determine the parameter adjustment order of the hyperparameters of the network indicator and the candidate values ​​of each of the hyperparameters; the pre-stored corresponding relationship information is the corresponding information between the time series feature category and the preset template, and the template includes the parameter adjustment order and the candidate values; The step of obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters includes: According to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters, obtaining the effect parameter value of the network indicator corresponding to each candidate value of the current hyperparameter to be adjusted; According to the effect parameter value, an initial target value of the current hyperparameter to be adjusted is determined from various candidate values ​​of the current hyperparameter to be adjusted.

2. The information processing method according to claim 1, characterized in that: After updating the current value of the hyperparameter to be adjusted according to the initial target value, the method further includes: Obtaining the impact amplitude value corresponding to each of the hyperparameters; According to the impact amplitude value, updating the parameter adjustment sequence; According to the updated parameter adjustment order, return to execute the parameter adjustment order, determine the current hyperparameter to be adjusted from each hyperparameter; obtain the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; update the current value of the current hyperparameter to be adjusted according to the initial target value.

3. The information processing method according to claim 2, characterized in that: The obtaining of the influence amplitude value corresponding to each of the hyperparameters includes: Obtaining an influence amplitude value corresponding to the current hyperparameter to be adjusted according to an effect parameter value corresponding to the current value of the hyperparameter to be adjusted before being updated and an effect parameter value corresponding to the current value of the hyperparameter to be adjusted after being updated; According to the influence amplitude value corresponding to the current hyperparameter to be adjusted, the influence amplitude value corresponding to each hyperparameter is obtained.

4. The information processing method according to claim 1, characterized in that: After updating the current value of the hyperparameter to be adjusted according to the initial target value, the method further includes: Determine a group of first sampling points according to the initial target value, the adjustment step size and the number of first sampling points; the first sampling points are located on both sides of the initial target value; Obtaining the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value; According to the first optimal effect parameter value and the first worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

5. The information processing method according to claim 4, characterized in that: The step of determining the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value includes: When the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is less than or equal to the first threshold, the adjustment step is increased according to a preset amplitude, and the step of determining a group of first sampling points according to the initial target value and the adjustment step is returned according to the increased adjustment step; until the adjustment step reaches an upper limit of the step, a target effect parameter value is obtained from each effect parameter value corresponding to each of the first sampling points, and the value of the first sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

6. The information processing method according to claim 4 or 5, characterized in that: The step of determining the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value includes: When the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is greater than a first threshold, determining a search step length, a search direction, and a search endpoint; wherein the search step length is equal to the adjustment step length, and the search endpoint is a first sampling point at the end of the search direction; According to the search step size, search direction and search endpoint, a final target value of the current hyperparameter to be adjusted is determined.

7. The information processing method according to claim 6, characterized in that: Determine the search direction, including: A search direction is determined according to the initial target value and the first sampling point corresponding to the optimal effect parameter value.

8. The information processing method according to claim 6, characterized in that: Determining the final target value of the current hyperparameter to be adjusted according to the search step size, the search direction, and the search endpoint includes: Determine a set of second sampling points according to the search step size, the search direction, the search endpoint, and the number of second sampling points; Obtaining the effect parameter value corresponding to each of the second sampling points to obtain a second optimal effect parameter value and a second worst effect parameter value; According to the second optimal effect parameter value and the second worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

9. The information processing method according to claim 8, characterized in that: The step of determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to the third threshold, re-determine the search step size, search direction and search endpoint; According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint; Among them, the re-determined search direction is consistent with the last search direction; The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; The search endpoint is a second sampling point located at the end of the search direction; The second threshold is smaller than the first threshold.

10. The information processing method according to claim 8, characterized in that: The step of determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, determine that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted; The second threshold is smaller than the first threshold.

11. The information processing method according to claim 8, characterized in that: The step of determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

12. The information processing method according to claim 8, characterized in that: The step of determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: When the absolute value of the difference between the second best effect parameter value and the second worst effect parameter value is greater than a second threshold, re-determining the search direction; Determining a final target value of the current hyperparameter to be adjusted according to the re-determined search direction; The re-determined search direction is the direction from the second sampling point corresponding to the second worst effect parameter value to the second sampling point corresponding to the second best effect parameter value; The second threshold is smaller than the first threshold.

13. The information processing method according to claim 12, characterized in that: Determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: When the re-determined search direction is consistent with the last search direction, or the re-determined search direction is not consistent with the last search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to a fourth threshold, re-determine the search step size and the search endpoint; According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint; The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; The search endpoint is a second sampling point located at the end of the search direction; The second threshold is smaller than the first threshold.

14. The information processing method according to claim 12, characterized in that: Determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: When the re-determined search direction is inconsistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the fourth threshold, the value of the sampling point corresponding to the second optimal effect parameter value is determined to be the final target value of the current hyperparameter to be adjusted.

15. The information processing method according to claim 12, characterized in that: Determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: When the re-determined search direction is consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

16. An information processing device, characterized in that: include: A first determination module, used to determine the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the time series feature category information of the network indicators; A second determination module is used to determine the current hyperparameter to be adjusted from each of the hyperparameters according to the parameter adjustment order; A first processing module, configured to obtain an initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; A first updating module, used for updating a current value of the hyperparameter to be adjusted according to the initial target value; The other hyperparameters are the hyperparameters of the network indicators except the hyperparameters to be adjusted currently. The information processing device further includes: A third determination module is used to determine the timing characteristic category information of the network indicator according to the timing characteristic parameter information of the network indicator before determining the parameter adjustment order of the hyperparameters of the network indicator and the candidate values ​​of each of the hyperparameters according to the timing characteristic category information of the network indicator; wherein the timing characteristic parameter information includes: at least one of indicator direction information, timing characteristic information, value type information and value magnitude information; The step of determining the adjustment order of the hyperparameters of the network indicators and the candidate values ​​of the hyperparameters according to the time series feature category information of the network indicators includes: According to the time series feature category information of the network indicator and the pre-stored corresponding relationship information, determine the parameter adjustment order of the hyperparameters of the network indicator and the candidate values ​​of each of the hyperparameters; the pre-stored corresponding relationship information is the corresponding information between the time series feature category and the preset template, and the template includes the parameter adjustment order and the candidate values; The first processing module comprises: A first processing submodule, configured to obtain, according to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters, an effect parameter value of the network indicator corresponding to each candidate value of the current hyperparameter to be adjusted; The first determination submodule is used to determine the initial target value of the current hyperparameter to be adjusted from various candidate values ​​of the current hyperparameter to be adjusted according to the effect parameter value.

17. The information processing device according to claim 16, characterized in that: Also includes: A first acquisition module, configured to acquire the impact amplitude value corresponding to each of the hyperparameters after updating the current value of the hyperparameter to be adjusted according to the initial target value; A second updating module, used for updating the parameter adjustment sequence according to the impact amplitude value; A second processing module is used to return to execute the step of determining the current hyperparameter to be adjusted from each hyperparameter according to the parameter adjustment sequence after the update; and obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; Update the current value of the hyperparameter to be adjusted according to the initial target value.

18. The information processing device according to claim 17, characterized in that: The first acquisition module includes: A second processing submodule is used to obtain the influence amplitude value corresponding to the current hyperparameter to be adjusted according to the effect parameter value corresponding to the current value of the hyperparameter to be adjusted before the current value is updated and the effect parameter value corresponding to the current value of the hyperparameter to be adjusted after the current value is updated; The third processing submodule is used to obtain the influence amplitude value corresponding to each of the hyperparameters according to the influence amplitude value corresponding to the current hyperparameter to be adjusted.

19. The information processing device according to claim 16, characterized in that: Also includes: a fourth determination module, configured to determine a set of first sampling points according to the initial target value, the adjustment step size, and the number of first sampling points after updating the current value of the hyperparameter to be adjusted according to the initial target value; The first sampling points are located on both sides of the initial target value; A second acquisition module is used to acquire the effect parameter value corresponding to each of the first sampling points to obtain a first optimal effect parameter value and a first worst effect parameter value; The fifth determination module is used to determine the final target value of the current hyperparameter to be adjusted according to the first optimal effect parameter value and the first worst effect parameter value.

20. The information processing device according to claim 19, characterized in that: The fifth determining module comprises: The fourth processing submodule is used to increase the adjustment step according to a preset amplitude when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is less than or equal to the first threshold value, and return to execute the step of determining a group of first sampling points according to the initial target value and the adjustment step according to the increased adjustment step; until the adjustment step reaches the step upper limit, the target effect parameter value is obtained from the effect parameter values ​​corresponding to the first sampling points, and the value of the first sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

21. The information processing device according to claim 19 or 20, characterized in that: The fifth determining module comprises: A second determination submodule is used to determine a search step length, a search direction, and a search endpoint when the absolute value of the difference between the first optimal effect parameter value and the first worst effect parameter value is greater than a first threshold; wherein the search step length is equal to the adjustment step length, and the search endpoint is a first sampling point located at the end of the search direction; The third determination submodule is used to determine the final target value of the current hyperparameter to be adjusted according to the search step size, search direction and search endpoint.

22. The information processing device according to claim 21, characterized in that: Determine the search direction, including: A search direction is determined according to the initial target value and the first sampling point corresponding to the optimal effect parameter value.

23. The information processing device according to claim 21, characterized in that: Determining the final target value of the current hyperparameter to be adjusted according to the search step size, the search direction, and the search endpoint includes: Determine a set of second sampling points according to the search step size, the search direction, the search endpoint, and the number of second sampling points; Obtaining the effect parameter value corresponding to each of the second sampling points to obtain a second optimal effect parameter value and a second worst effect parameter value; According to the second optimal effect parameter value and the second worst effect parameter value, a final target value of the current hyperparameter to be adjusted is determined.

24. The information processing device according to claim 23, characterized in that: The step of determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to the third threshold, re-determine the search step size, search direction and search endpoint; According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint; Among them, the re-determined search direction is consistent with the last search direction; The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; The search endpoint is a second sampling point located at the end of the search direction; The second threshold is smaller than the first threshold.

25. The information processing device according to claim 23, characterized in that: The step of determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, determine that the value of the sampling point corresponding to the second optimal effect parameter value is the final target value of the current hyperparameter to be adjusted; The second threshold is smaller than the first threshold.

26. The information processing device according to claim 23, characterized in that The step of determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: When the absolute value of the difference between the second optimal effect parameter value and the second worst effect parameter value is less than or equal to the second threshold, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the third threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

27. The information processing device according to claim 23, characterized in that: The step of determining the final target value of the current hyperparameter to be adjusted according to the second optimal effect parameter value and the second worst effect parameter value includes: When the absolute value of the difference between the second best effect parameter value and the second worst effect parameter value is greater than a second threshold, re-determining the search direction; Determining a final target value of the current hyperparameter to be adjusted according to the re-determined search direction; The re-determined search direction is the direction from the second sampling point corresponding to the second worst effect parameter value to the second sampling point corresponding to the second best effect parameter value; The second threshold is smaller than the first threshold.

28. The information processing device according to claim 27, characterized in that: Determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: When the re-determined search direction is consistent with the last search direction, or the re-determined search direction is not consistent with the last search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is greater than or equal to a fourth threshold, re-determine the search step size and the search endpoint; According to the re-determined search step size, search direction, and search endpoint, returning to execute the step of determining the final target value of the current hyperparameter to be adjusted according to the search step size, search direction, and search endpoint; The search step length is equal to the absolute value of the difference between the second sampling point corresponding to the second optimal effect parameter value and the second sampling point corresponding to the second worst effect parameter value; The search endpoint is a second sampling point located at the end of the search direction; The second threshold is smaller than the first threshold.

29. The information processing device according to claim 27, characterized in that: Determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: When the re-determined search direction is inconsistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than the fourth threshold, the value of the sampling point corresponding to the second optimal effect parameter value is determined to be the final target value of the current hyperparameter to be adjusted.

30. The information processing device according to claim 27, characterized in that: Determining the final target value of the current hyperparameter to be adjusted according to the re-determined search direction includes: When the re-determined search direction is consistent with the previous search direction, and the absolute value of the difference between the second optimal effect parameter value and the optimal effect parameter value corresponding to the sampling points of the previous batch is less than a fourth threshold, the target effect parameter value is obtained from the effect parameter values ​​corresponding to each of the first sampling points and the effect parameter values ​​corresponding to each of the second sampling points, and the value of the sampling point corresponding to the target effect parameter value is used as the final target value of the current hyperparameter to be adjusted.

31. An information processing device, characterized in that: include: processor; The processor is used to determine the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the time series feature category information of the network indicators; According to the parameter adjustment order, determine the current hyperparameter to be adjusted from each of the hyperparameters; Obtaining an initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters; Updating the current value of the current hyperparameter to be adjusted according to the initial target value; The other hyperparameters are the hyperparameters of the network indicators except the hyperparameters to be adjusted currently. The processor is further configured to: Before determining the parameter adjustment order of the hyperparameters of the network indicators and the candidate values ​​of each of the hyperparameters according to the timing characteristic category information of the network indicators, determining the timing characteristic category information of the network indicators according to the timing characteristic parameter information of the network indicators; wherein the timing characteristic parameter information includes: at least one of indicator direction information, timing characteristic information, value type information, and value magnitude information; The step of determining the adjustment order of the hyperparameters of the network indicators and the candidate values ​​of the hyperparameters according to the time series feature category information of the network indicators includes: According to the time series feature category information of the network indicator and the pre-stored corresponding relationship information, determine the parameter adjustment order of the hyperparameters of the network indicator and the candidate values ​​of each of the hyperparameters; the pre-stored corresponding relationship information is the corresponding information between the time series feature category and the preset template, and the template includes the parameter adjustment order and the candidate values; The step of obtaining the initial target value of the current hyperparameter to be adjusted according to the candidate value of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters includes: According to the candidate values ​​of the current hyperparameter to be adjusted and the current values ​​of other hyperparameters, obtaining the effect parameter value of the network indicator corresponding to each candidate value of the current hyperparameter to be adjusted; According to the effect parameter value, an initial target value of the current hyperparameter to be adjusted is determined from various candidate values ​​of the current hyperparameter to be adjusted.

32. An information processing device, comprising a memory, a processor, and a program stored in the memory and executable on the processor; characterized in that: When the processor executes the program, the information processing method according to any one of claims 1 to 15 is implemented.

33. A readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the information processing method according to any one of claims 1 to 15 are implemented.

Images