Model training method based on computing power of intelligent computing center and related equipment

Abstract

The invention provides a model training method based on computing power of an intelligent computing center and related equipment, and relates to the technical field of computing power infrastructure, and the method comprises the steps: S1, obtaining initial data and a plurality of preset features corresponding to the initial data; s2, deriving the plurality of preset features to obtain a plurality of derived features; s3, screening from the plurality of preset features and the plurality of derivative features based on the initial data to obtain N first features, wherein N is a positive integer greater than 1; and S4, a preset model is trained based on the N first features, a target model is obtained, and the target model is used for processing data based on the N first features. According to the method, the data value can be fully mined, the model effect can be improved, and meanwhile, the model training efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computing power infrastructure, and in particular to a model training method and related equipment based on the computing power of an intelligent computing center. Background Art

[0002] "Intelligent computing center" refers to a facility that provides the required computing power, data and algorithms for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning) by using large-scale heterogeneous computing resources, including general computing power and intelligent computing power. The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0003] In the modeling process of machine learning, it is necessary to generate feature variables through feature engineering, and then train the model with feature variables and sample data to obtain a model that meets the requirements.

[0004] At present, algorithm personnel usually manually derive feature variables based on their understanding of business scenarios. However, manually generated feature variables are limited by the business experience of algorithm personnel, and can only be based on features that can be imagined within the scope of experience. Features that are effective for model prediction outside of experience cannot be obtained, and the value of data cannot be fully explored. Moreover, it is difficult to generate feature variables based on business experience in batches, resulting in low efficiency in generating features, which in turn affects the efficiency of model training.

[0005] It can be seen that the related technology has the problem of generating features based on experience, which leads to failure to fully explore the value of data and low training efficiency. Summary of the invention

[0006] The embodiment of the present invention provides a model training method and related equipment based on the computing power of an intelligent computing center to solve the problem in the related art that the efficiency of generating features is low, resulting in low model training efficiency.

[0007] To solve the above problems, the present invention is achieved as follows:

[0008] In a first aspect, an embodiment of the present invention provides a model training method based on the computing power of an intelligent computing center, comprising:

[0009] Step S1, obtaining initial data and a plurality of preset features corresponding to the initial data;

[0010] Step S2, deriving the plurality of preset features to obtain a plurality of derived features;

[0011] Step S3, based on the initial data, screening out N first features from the plurality of preset features and the plurality of derived features, where N is a positive integer greater than 1;

[0012] Step S4: training a preset model based on the N first features to obtain a target model, wherein the target model is used to process data based on the N first features.

[0013] In one embodiment, step S3 includes:

[0014] Step S31, calculating the evaluation parameter corresponding to each of the plurality of preset features and the plurality of derived features based on the initial data;

[0015] Step S32: based on the preset conditions and the evaluation parameters corresponding to the features, screen from the multiple preset features and the multiple derived features to obtain the N first features, wherein the evaluation parameters corresponding to each of the N first features satisfy the preset conditions, and the preset conditions are used to characterize the relevant information of the first feature.

[0016] In one embodiment, the evaluation parameters include at least one of missing rate, relevance, information value IV value, population stability index PSI value, and single value rate;

[0017] The preset condition includes at least one of the following:

[0018] The missing rate corresponding to the first feature is less than a set missing rate threshold;

[0019] The correlation corresponding to the first feature is less than a set correlation threshold;

[0020] The IV value corresponding to the first feature is greater than a set IV threshold;

[0021] The PSI value corresponding to the first feature is greater than a set PSI threshold;

[0022] The single value rate corresponding to the first feature is less than a set single value rate threshold.

[0023] In one embodiment, step S4 includes:

[0024] Step S41: training the preset model based on the N first features to obtain a first intermediate model and a first loss value corresponding to the first intermediate model;

[0025] Step S42: When the first loss value is less than or equal to a set loss threshold, the first intermediate model is set as the target model.

[0026] In one embodiment, step S4 further includes:

[0027] Step S43: when the first loss value is greater than the set loss threshold, based on the initial data, screen out N second features from the plurality of preset features and the plurality of derived features, wherein the N second features are different from the N first features in at least one feature;

[0028] Step S44: training the preset model based on the N second features to obtain a second intermediate model and a second loss value corresponding to the second intermediate model;

[0029] Step S45, repeating step S43 and step S44 until the second loss value is less than or equal to the set loss threshold, setting the second intermediate model as the target model.

[0030] In one embodiment, the number of preset features in the N first features accounts for a proportion greater than or equal to a set ratio.

[0031] In a second aspect, an embodiment of the present invention further provides a model training device based on the computing power of an intelligent computing center, comprising:

[0032] An acquisition module, used to acquire initial data and a plurality of preset features corresponding to the initial data;

[0033] A derivation module, used for deriving the plurality of preset features to obtain a plurality of derived features;

[0034] A screening module, configured to screen the plurality of preset features and the plurality of derived features based on the initial data to obtain N first features, where N is a positive integer greater than 1;

[0035] A training module is used to train a preset model based on the N first features to obtain a target model, and the target model is used to process data based on the N first features.

[0036] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps in the model training method based on the computing power of an intelligent computing center as described in the first aspect above are implemented.

[0037] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the model training method based on the computing power of an intelligent computing center as described in the first aspect above are implemented.

[0038] In a fifth aspect, an embodiment of the present invention further provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps in the model training method based on the computing power of an intelligent computing center as described in the first aspect above.

[0039] In an embodiment of the present invention, multiple preset features and initial data corresponding to the multiple preset features are obtained; the multiple preset features are derived to obtain multiple derived features; based on the initial data, N first features are screened from the multiple preset features and the multiple derived features, where N is a positive integer greater than 1; the preset model is trained based on the N first features to obtain a target model, and the target model is used to process data based on the N first features. In this way, by deriving multiple preset features to obtain multiple derived features, the limitations of business experience are broken, and derived features that cannot be obtained through business experience are added, so that the value of data is fully mined, and the efficiency of generating features is also improved; then, N first features are screened from the multiple preset features and the multiple derived features, and the N first features include features in multiple derived features. The quality of the obtained N first features is improved by filtering, and the N first features include the derived derived features. Finally, the preset model is trained based on the N first features, and the derived features as a supplement can improve the effect of model prediction, and the features generated quickly improve the training efficiency of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0041] Figure 1 It is a flow chart of a model training method based on the computing power of an intelligent computing center provided by an embodiment of the present invention;

[0042] Figure 2 It is a schematic diagram of the time of generating characteristics in the manual mode and the automatic mode provided by the embodiment of the present invention;

[0043] Figure 3 It is a schematic diagram of the number of features generated by the manual method and the automatic method provided in the embodiment of the present invention;

[0044] Figure 4 It is a schematic diagram of the quality of the generated features in the manual method and the automatic method provided by the embodiment of the present invention;

[0045] Figure 5is a schematic diagram of the importance scoring of features provided by an embodiment of the present invention;

[0046] Figure 6 Schematic diagram of ROC of different models provided by embodiments of the present invention;

[0047] Figure 7 It is a structural diagram of a model training device based on the computing power of an intelligent computing center provided by an embodiment of the present invention;

[0048] Figure 8 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The "computing power" mentioned in the present invention is the ability of computer equipment or computing / data center to process information. It is the ability of computer hardware and software to work together to execute certain computing requirements. It is the computing power to achieve target result output by processing information data. It is a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to the society through computing power infrastructure.

[0051] The "computing power" (Computational Power, CP) described in the present invention is the ability of a data center server to process data and output results. It is a comprehensive indicator to measure the computing power of a data center, including general computing power, super computing power and intelligent computing power. The commonly used unit of measurement is the number of floating point operations performed per second (FLOPS: Floating Point Operations Per Second, 1EFLOPS = 10^18FLOPS). The larger the value, the stronger the comprehensive computing power. According to calculations, 1EFLOPS is approximately the computing power output of 5 Tianhe-2A or 500,000 mainstream server CPUs or 2 million mainstream notebooks. The calculation formula is: CP = CP 通用 +CP 智能 +CP 超级 .

[0052] The "Network Power" (NP) described in the present invention is a manifestation of the data transmission capability of computing facilities, including comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, etc. The carrying capacity involves network transmission within and between data centers, and is a comprehensive indicator for measuring network transmission scheduling capabilities. In the embodiment of the present invention, the carrying capacity uses the memory bandwidth.

[0053] The "Storage Power" (SP) described in the present invention is the comprehensive ability of a data center in terms of data storage capacity, performance, safety and reliability, and green and low-carbon. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and server built-in storage devices. The commonly used unit of measurement for storage capacity is exabyte (EB, 1EB = 2^60bytes), and the commonly used unit of measurement for performance is the number of reads and writes per second per unit capacity (IOPS / TB, Input / Output Operations Per Second / TB). The disaster recovery ratio is an important manifestation of safety and reliability.

[0054] The "computing power infrastructure" described in the present invention is a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity. It can realize centralized computing, storage, transmission and application of information, and presents characteristics such as diversity and ubiquity, intelligence and agility, safety and reliability, green and low-carbon. It is of great significance to promote industrial transformation and upgrading, enable my country's scientific and technological innovation, meet people's better life and realize high-efficiency social governance.

[0055] The “computing power” mentioned in the present invention includes general computing power, intelligent computing power and super computing power.

[0056] The “general computing power” mentioned in the present invention refers to the computing power provided by a server based on a central processing unit (CPU) chip, which is used to support basic general computing such as cloud computing and edge computing.

[0057] The "intelligent computing power" described in the present invention is a computing platform for large-scale deployment of special chips such as graphics processing units (GPUs), field programmable gate arrays (FPGAs), and application-specific integrated circuits (ASICs) for various innovative applications of artificial intelligence, such as natural language processing and machine vision.

[0058] The "super computing power" mentioned in the present invention is mainly the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for calculations in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0059] The "intelligent computing center" mentioned in the present invention refers to a facility that provides the required computing power, data and algorithms for artificial intelligence applications (such as artificial intelligence deep learning model development, model training and model reasoning) by using large-scale heterogeneous computing resources, including general computing power (CPU) and intelligent computing power (GPU, FPGA, etc.). The intelligent computing center covers facilities, hardware, and software, and can provide full-stack capabilities from bottom-level computing power to top-level application enablement.

[0060] The "computing resources" mentioned in the present invention refer to the technologies and facilities with information calculation, transmission, storage and application capabilities required for the development of the digital society, including but not limited to computing resources such as CPUs and GPUs, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and supporting and guarantee resources such as wind, fire, water and electricity.

[0061] The "features" mentioned in the present invention refer to the features imported into the model during the model training process of machine learning. The training effect of the model is adjusted by importing different features to obtain models for different application scenarios. Among them, features can include user-related features, product-related features, and scenario-related features.

[0062] Among them, the "preset features" described in the present invention are feature variables manually generated by business personnel based on business experience and testing. It can be understood that the number of preset features is set according to business needs, and since the preset features are feature variables obtained by business personnel based on business experience and testing, the preset features have better interpretability. In addition, the "derived features" described in the present invention are new features derived from preset features. Derived features are randomly generated. Compared with preset features, derived features have poorer interpretability.

[0063] The "derivation" mentioned in the present invention refers to combining at least two features to obtain a new derived feature. The combination of features can be a simple aggregation, such as by calculating the sum, difference or average value corresponding to at least two features, or a deep feature synthesis method through a preset function, for example, by an automatic method such as feature tools (FeatureTools) to synthesize new derived features.

[0064] The “preset model” described in the present invention is an initial model without imported features, and the “target model” described in the present invention is a model that imports features and is trained based on the features. The target model can be used to analyze data based on the imported features.

[0065] See also Figure 1 , Figure 1 is a flow chart of a model training method based on the computing power of an intelligent computing center provided by an embodiment of the present invention. Figure 1 As shown, the following steps are included:

[0066] Step S1, obtaining initial data and a plurality of preset features corresponding to the initial data;

[0067] Step S2, deriving the plurality of preset features to obtain a plurality of derived features;

[0068] Step S3, based on the initial data, screening out N first features from the plurality of preset features and the plurality of derived features, where N is a positive integer greater than 1;

[0069] Step S4: training a preset model based on the N first features to obtain a target model, wherein the target model is used to process data based on the N first features.

[0070] In an embodiment of the present invention, multiple preset features and initial data corresponding to the multiple preset features are obtained; the multiple preset features are derived to obtain multiple derived features; based on the initial data, N first features are screened from the multiple preset features and the multiple derived features, where N is a positive integer greater than 1; the preset model is trained based on the N first features to obtain a target model, and the target model is used to process data based on the N first features. In this way, by deriving multiple preset features to obtain multiple derived features, the limitations of business experience are broken, and derived features that cannot be obtained through business experience are added, so that the value of data is fully mined, and the efficiency of generating features is also improved; then, N first features are screened from the multiple preset features and the multiple derived features, and the N first features include features in multiple derived features. The quality of the obtained N first features is improved by filtering, and the N first features include the derived derived features. Finally, the preset model is trained based on the N first features, and the derived features as a supplement can improve the effect of model prediction, and the features generated quickly improve the training efficiency of the model.

[0071] It should be noted that there is a significant difference between the efficiency of manual methods, where business personnel obtain feature variables based on business experience and testing, and the efficiency of deriving features through automatic methods. Deriving features through automatic methods can effectively improve the efficiency of generating features. Figure 2 and Figure 3 As shown in the figure, when the time consumed by the manual method is 10 times that of the automatic method, the number of features obtained by the manual method is much less than the number of features derived by the automatic method. Based on this, deriving multiple derived features from multiple preset features can effectively improve the efficiency of generating features, thereby improving the efficiency of model training.

[0072] In one embodiment, step S3 includes:

[0073] Step S31, calculating the evaluation parameter corresponding to each of the plurality of preset features and the plurality of derived features based on the initial data;

[0074] Step S32: based on the preset conditions and the evaluation parameters corresponding to the features, screen from the multiple preset features and the multiple derived features to obtain the N first features, wherein the evaluation parameters corresponding to each of the N first features satisfy the preset conditions, and the preset conditions are used to characterize the relevant information of the first feature.

[0075] It should be noted that the qualities of multiple preset features and multiple derived features derived from multiple preset features are different. Some features cannot achieve the expected results after being imported into the model. It is necessary to filter out these features and obtain the most expected N first features to train the model, so that the trained model can accurately analyze the data.

[0076] For example, Figure 4 As shown in the figure, the quality of the preset features obtained manually is similar to that of the derived features, and both can improve the model effect to a certain extent. However, the improvement effect shows that not all preset features or derived features have a positive effect on model training. It is necessary to filter out some invalid features and retain only some valid features to improve the analysis effect of the trained model.

[0077] In an embodiment of the present invention, the evaluation parameters corresponding to each of the multiple preset features and the multiple derived features are calculated based on the initial data; the multiple preset features and the multiple derived features are screened based on the preset conditions and the evaluation parameters corresponding to the features to obtain the N first features, and the evaluation parameters corresponding to each of the N first features meet the preset conditions, and the preset conditions are used to characterize the relevant information of the first features. In this way, all the preset features and derived features are screened by the preset conditions to obtain N first features, thereby improving the analysis effect of the target data obtained by training based on the N first features.

[0078] In one embodiment, the evaluation parameters include at least one of missing rate, relevance, information value (IV) value, population stability index (PSI) value, and single value rate;

[0079] The preset condition includes at least one of the following:

[0080] The missing rate corresponding to the first feature is less than a set missing rate threshold;

[0081] The correlation corresponding to the first feature is less than a set correlation threshold;

[0082] The IV value corresponding to the first feature is greater than a set IV threshold;

[0083] The PSI value corresponding to the first feature is greater than a set PSI threshold;

[0084] The single value rate corresponding to the first feature is less than a set single value rate threshold.

[0085] The above missing rate is used to characterize the actual situation of the value in the initial data corresponding to the feature. For example, if a feature corresponds to 10 users, but only 9 users have data, the missing rate corresponding to the feature is 10%. It should be noted that the higher the missing rate, the less data can be analyzed, and the worse the training effect after the feature is imported into the model. The missing rate needs to be controlled within the set missing rate threshold.

[0086] The above correlation is used to characterize the similarity between the feature and other features. It should be noted that the higher the correlation between two features, the closer the effects of the two features. The effect of importing two highly correlated features into model training is close to the effect of importing one of the features. In order to import more features, it is necessary to filter the two highly correlated features through the correlation threshold.

[0087] The above IV value is used to indicate the contribution of the feature to the target prediction, that is, the predictive ability of the feature. Generally speaking, the higher the IV value, the stronger the predictive ability of the feature, and the higher the information contribution. In the embodiment of the present invention, in order to improve the predictive ability of the model obtained by training the first feature, it is necessary to set the IV threshold to screen the corresponding features with high IV values, thereby improving the predictive ability of the target model.

[0088] The above PSI value is used to measure the stability of the feature in time distribution. The higher the PSI value, the higher the stability of the feature in time distribution, and the higher the stability of the trained model in time distribution. In an embodiment of the present invention, in order to improve the stability of the trained model in time distribution, it is necessary to set a PSI threshold to screen the features, thereby improving the stability of the target model in time distribution.

[0089] The above single value rate is used to characterize the proportion of a variable in all values ​​in the data corresponding to the feature. The higher the single value rate, the less the data corresponding to the feature changes. It is necessary to filter out features with high single value rates through the single value rate threshold to improve the effect of the target model on data analysis.

[0090] In the embodiment of the present invention, the evaluation parameters include at least one of missing rate, correlation, IV value, PSI value, and single value rate; the preset conditions include at least one of the following: the missing rate corresponding to the first feature is less than the set missing rate threshold; the correlation corresponding to the first feature is less than the set correlation threshold; the IV value corresponding to the first feature is greater than the set IV threshold; the PSI value corresponding to the first feature is greater than the set PSI threshold; the single value rate corresponding to the first feature is less than the set single value rate threshold. In this way, all features are filtered through the evaluation parameters and preset conditions to improve the quality of the N first features obtained by screening.

[0091] In an optional embodiment, when there are more than N features whose corresponding evaluation parameters meet preset conditions, at least one of the missing rate, correlation, IV value, PSI value, and single value rate corresponding to the features can be weighted to obtain the importance score of the feature, and then the N first features are obtained by screening through the importance score.

[0092] For example, Figure 5 As shown, 15 features are obtained by screening through importance scores, and the importance scores of the features are sorted from high to low. Some features can be screened from the 15 features as the first features.

[0093] In one embodiment, step S4 includes:

[0094] Step S41: training the preset model based on the N first features to obtain a first intermediate model and a first loss value corresponding to the first intermediate model;

[0095] Step S42: When the first loss value is less than or equal to a set loss threshold, the first intermediate model is set as the target model.

[0096] It should be noted that the first features obtained by screening through evaluation parameters and preset conditions may also contain some features with poor effects, which need to be further screened. At this time, the screening is mainly achieved through the training effect of the model.

[0097] In an embodiment of the present invention, a preset model is trained based on N first features to obtain a first intermediate model and a first loss value corresponding to the first intermediate model; when the first loss value is less than or equal to a set loss threshold, the first intermediate model is set as a target model, thereby determining the target model by setting the loss threshold.

[0098] In one embodiment, step S4 further includes:

[0099] Step S43: when the first loss value is greater than the set loss threshold, based on the initial data, screen out N second features from the plurality of preset features and the plurality of derived features, wherein the N second features are different from the N first features in at least one feature;

[0100] Step S44: training the preset model based on the N second features to obtain a second intermediate model and a second loss value corresponding to the second intermediate model;

[0101] Step S45, repeating step S43 and step S44 until the second loss value is less than or equal to the set loss threshold, setting the second intermediate model as the target model.

[0102] In the embodiment of the present invention, a loss threshold is set to determine whether the training effect of the model has reached expectations. If the corresponding loss value after model training is less than or equal to the set loss threshold, it is considered that the expectation has been reached, and if the corresponding loss value after model training is greater than the set loss threshold, it is considered that the expectation has not been reached, and it is necessary to re-screen the N second features to improve the training effect.

[0103] In an optional embodiment, re-screening the N second features may specifically be replacing some of the N first features with other features, for example, replacing multiple features with the lowest importance scores among the N first features with multiple features with the highest importance scores among the other features.

[0104] In one embodiment, the number of preset features in the N first features accounts for a proportion greater than or equal to a set ratio.

[0105] In an embodiment of the present invention, the interpretability of preset features is better than that of derived features. In order to ensure the interpretability of the target model finally trained, it is necessary to control the number of derived features in the N first features within a certain range, that is, the proportion of preset features in the N first features is greater than or equal to the set ratio.

[0106] The embodiment of the present invention also provides a comparison diagram of a first model trained only by preset features for sales data of different months, a target model trained by N first features, and a receiver operating characteristic curve (ROC) of a baseline, as shown in FIG. Figure 6 As shown in the figure, it can be seen that the target model trained by N first features has more accurate prediction.

[0107] See also Figure 7 , Figure 7 is a structural diagram of a model training device based on intelligent computing center computing power provided by an embodiment of the present invention, such as Figure 7 As shown, the model training device 700 based on the computing power of the intelligent computing center includes:

[0108] An acquisition module 701 is used to acquire initial data and a plurality of preset features corresponding to the initial data;

[0109] A derivation module 702, configured to derive the plurality of preset features to obtain a plurality of derived features;

[0110] A screening module 703 is used to screen the plurality of preset features and the plurality of derived features based on the initial data to obtain N first features, where N is a positive integer greater than 1;

[0111] The training module 704 is used to train the preset model based on the N first features to obtain a target model, and the target model is used to process data based on the N first features.

[0112] In one embodiment, the screening module 703 includes:

[0113] An evaluation unit, configured to calculate an evaluation parameter corresponding to each of the plurality of preset features and the plurality of derived features based on the initial data;

[0114] The first screening unit is used to screen from the multiple preset features and the multiple derived features based on preset conditions and evaluation parameters corresponding to the features to obtain the N first features, where the evaluation parameter corresponding to each first feature in the N first features meets the preset conditions, and the preset conditions are used to characterize relevant information of the first feature.

[0115] In one embodiment, the evaluation parameters include at least one of missing rate, relevance, information value IV value, population stability index PSI value, and single value rate;

[0116] The preset condition includes at least one of the following:

[0117] The missing rate corresponding to the first feature is less than a set missing rate threshold;

[0118] The correlation corresponding to the first feature is less than a set correlation threshold;

[0119] The IV value corresponding to the first feature is greater than a set IV threshold;

[0120] The PSI value corresponding to the first feature is greater than a set PSI threshold;

[0121] The single value rate corresponding to the first feature is less than a set single value rate threshold.

[0122] In one embodiment, the training module 704 includes:

[0123] A first training unit, configured to train the preset model based on the N first features to obtain a first intermediate model and a first loss value corresponding to the first intermediate model;

[0124] The first setting unit is used to set the first intermediate model as the target model when the first loss value is less than or equal to a set loss threshold.

[0125] In one embodiment, the training module 704 further includes:

[0126] a second screening unit, configured to screen, based on the initial data, N second features from the plurality of preset features and the plurality of derived features when the first loss value is greater than the set loss threshold, wherein at least one feature of the N second features is different from that of the N first features;

[0127] A second training unit, used for training the preset model based on the N second features to obtain a second intermediate model and a second loss value corresponding to the second intermediate model;

[0128] The second setting unit is used to repeatedly screen N second features and train until the second loss value is less than or equal to the set loss threshold, and then set the second intermediate model as the target model.

[0129] In one embodiment, the number of preset features in the N first features accounts for a proportion greater than or equal to a set ratio.

[0130] The model training device based on the computing power of an intelligent computing center provided in an embodiment of the present invention is capable of implementing each process of each embodiment of the above-mentioned model training method based on the computing power of an intelligent computing center. The technical features correspond one to one and can achieve the same technical effect. To avoid repetition, they will not be described here.

[0131] It should be noted that the model training device based on the computing power of the intelligent computing center in the embodiment of the present invention can be a device, or it can be a component, integrated circuit, or chip in an electronic device.

[0132] The embodiment of the present invention further provides an electronic device, see Figure 8 , Figure 8 801, a processor 802, and a program or instruction stored in the memory 801 and executed by the processor 802. Figure 1 Any steps in the corresponding embodiment of the model training method based on the computing power of the intelligent computing center and the same beneficial effects are achieved will not be repeated here.

[0133] The processor 802 may be a CPU, an ASIC, an FPGA or a GPU.

[0134] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above-mentioned model training method embodiment based on the computing power of the intelligent computing center can be completed by hardware related to program instructions, and the program can be stored in a readable medium.

[0135] The embodiment of the present invention further provides a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above Figure 1 The corresponding steps in the embodiment of the model training method based on the computing power of the intelligent computing center can achieve the same technical effect. To avoid repetition, they are not repeated here. The storage medium is such as a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk.

[0136] The embodiment of the present invention also provides a computer program product, including computer instructions, which can achieve the above-mentioned Figure 1 Any steps in the corresponding implementation example of the model training method based on the computing power of the intelligent computing center can achieve the same technical effect and will not be repeated here to avoid repetition.

[0137] The terms "first", "second" etc. in the embodiments of the present invention are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. In addition, the terms "include" and "have" and any variation thereof are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment. In addition, "and / or" is used in the present application to represent at least one of the connected objects, such as A and / or B and / or C, indicating that A alone, B alone, C alone, and A and B all exist, B and C all exist, A and C all exist, and 7 situations in which A, B and C all exist.

[0138] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0139] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a second terminal device, etc.) to execute the methods of each embodiment of the present application.

[0140] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A model training method based on the computing power of an intelligent computing center, characterized in that: include: Step S1, obtaining initial data and a plurality of preset features corresponding to the initial data; Step S2, deriving the plurality of preset features to obtain a plurality of derived features; Step S3, based on the initial data, screening out N first features from the plurality of preset features and the plurality of derived features, where N is a positive integer greater than 1; Step S4: training a preset model based on the N first features to obtain a target model, wherein the target model is used to process data based on the N first features.

2. The method according to claim 1, characterized in that The step S3 comprises: Step S31, calculating the evaluation parameter corresponding to each of the plurality of preset features and the plurality of derived features based on the initial data; Step S32: based on the preset conditions and the evaluation parameters corresponding to the features, screen from the multiple preset features and the multiple derived features to obtain the N first features, wherein the evaluation parameters corresponding to each of the N first features satisfy the preset conditions, and the preset conditions are used to characterize the relevant information of the first feature.

3. The method according to claim 2, characterized in that The evaluation parameters include at least one of missing rate, correlation, information value IV value, population stability index PSI value, and single value rate; The preset condition includes at least one of the following: The missing rate corresponding to the first feature is less than a set missing rate threshold; The correlation corresponding to the first feature is less than a set correlation threshold; The IV value corresponding to the first feature is greater than a set IV threshold; The PSI value corresponding to the first feature is greater than a set PSI threshold; The single value rate corresponding to the first feature is less than a set single value rate threshold.

4. The method according to claim 1, characterized in that The step S4 comprises: Step S41: training the preset model based on the N first features to obtain a first intermediate model and a first loss value corresponding to the first intermediate model; Step S42: When the first loss value is less than or equal to a set loss threshold, the first intermediate model is set as the target model.

5. The method according to claim 4, characterized in that The step S4 further comprises: Step S43: when the first loss value is greater than the set loss threshold, based on the initial data, screen out N second features from the plurality of preset features and the plurality of derived features, wherein the N second features are different from the N first features in at least one feature; Step S44: training the preset model based on the N second features to obtain a second intermediate model and a second loss value corresponding to the second intermediate model; Step S45, repeating step S43 and step S44 until the second loss value is less than or equal to the set loss threshold, setting the second intermediate model as the target model.

6. The method according to claim 1, characterized in that The number of preset features in the N first features is greater than or equal to a set ratio.

7. A model training device based on the computing power of an intelligent computing center, characterized in that: include: An acquisition module, used to acquire initial data and a plurality of preset features corresponding to the initial data; A derivation module, used for deriving the plurality of preset features to obtain a plurality of derived features; A screening module, configured to screen the plurality of preset features and the plurality of derived features based on the initial data to obtain N first features, where N is a positive integer greater than 1; A training module is used to train a preset model based on the N first features to obtain a target model, and the target model is used to process data based on the N first features.

8. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the model training method based on the computing power of an intelligent computing center are implemented as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the model training method based on the computing power of an intelligent computing center as described in any one of claims 1 to 6.

10. A computer program product, characterized in that It includes computer instructions, which, when executed by a processor, implement the steps of the model training method based on the computing power of an intelligent computing center as described in any one of claims 1 to 6.

Images