Model training method and device, equipment, storage medium and program product
By analyzing the feature distribution parameters of the data set and deleting unnecessary features, combined with supervision and weak supervision training, the prediction inaccuracy problem caused by low data quality in the existing technology is solved, and the prediction accuracy and robustness of the model are improved.
Patent Information
- Application Number
- CN202411941586.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-04-22
AI Technical Summary
In the prior art, when machine learning and deep learning predict real data, the prediction results are inaccurate due to the low data quality, which is quite different from the actual results.
By obtaining the data set and splitting it into a training set and a verification set, the distribution parameters of the same feature in multiple data sets are determined, feature data that needs to be deleted, and supervised and weakly supervised training are carried out to improve the robustness and generalization ability of the model.
The quality of the data set is improved, the model prediction is more accurate, and the model's robustness to noise data and its predictive ability for unseen data is enhanced.
Smart Images
Figure CN119862420A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of big data, and in particular to a model training method, device, equipment, storage medium and program product. Background Art
[0002] The research on time series data is of far-reaching and extensive significance. It not only helps us understand the laws behind the data, but also provides a scientific basis for decision-making in various fields.
[0003] In the existing technology, machine learning and deep learning can adaptively learn the characteristics of data based on the input-output mapping of data, and then obtain an accurate prediction result. However, machine learning and deep learning are highly dependent on data. For ideal data with large data volume, clean characteristics and small deviation, machine learning and deep learning can obtain accurate prediction results.
[0004] However, the dimensions and features of real data are manually selected and labeled, so there are problems of insufficient data and low data quality. Therefore, in actual situations, the results obtained by predicting data through machine learning and deep learning are inaccurate and differ greatly from the actual results. Summary of the invention
[0005] The present application provides a model training method, device, equipment, storage medium and program product to solve the technical problem in the prior art that machine learning and deep learning make inaccurate predictions on data, which are significantly different from the actual results.
[0006] In a first aspect, the present application provides a model training method, comprising:
[0007] Acquire a first data set, and split the first data set into multiple second data sets, the multiple second data sets include: a training set and a validation set, the training set includes: a supervised training set and a weakly supervised training set, the supervised training set is a training set with accurate labels, and the weakly supervised training set is a training set without labels and / or with incorrect labels;
[0008] Determine distribution parameters of the same feature in multiple second data sets, and determine whether the feature is a feature to be deleted according to the multiple distribution parameters of each feature, each feature corresponds to one distribution parameter in each second data set;
[0009] If the feature is a feature to be deleted, deleting the data corresponding to the feature from the second data set;
[0010] Performing supervised training on the first prediction model according to the supervised training set to obtain a second prediction model, and performing weakly supervised training on the second prediction model according to the training set to obtain a third prediction model;
[0011] The performance of the third prediction model is verified using the verification set.
[0012] In a second aspect, the present application provides a model training device, comprising:
[0013] An acquisition module is used to acquire a first data set and split the first data set into multiple second data sets, wherein the multiple second data sets include: a training set and a validation set, wherein the training set includes: a supervised training set and a weakly supervised training set, wherein the supervised training set is a training set with accurate labels, and the weakly supervised training set is a training set with no labels and / or incorrect labels;
[0014] A processing module, used to determine distribution parameters of the same feature in multiple second data sets, and determine whether the feature is a feature to be deleted according to the multiple distribution parameters of each feature, each feature corresponding to one distribution parameter in each second data set;
[0015] The processing module is further configured to delete data corresponding to the feature from the second data set when the feature is a feature to be deleted;
[0016] The processing module is further used to perform supervised training on the first prediction model according to the supervised training set to obtain a second prediction model, and to perform weakly supervised training on the second prediction model according to the training set to obtain a third prediction model;
[0017] The processing module is further used to verify the performance of the third prediction model through the verification set.
[0018] In a third aspect, the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;
[0019] The memory stores computer-executable instructions;
[0020] The processor executes the computer-executable instructions stored in the memory to implement the method as described in any one of the first aspects.
[0021] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method as described in any one of the first aspects.
[0022] In a fifth aspect, the present application provides a computer program product, including a computer program, which implements the method described in any one of the first aspects when executed by a processor.
[0023] The present application provides a model training method, apparatus, device, storage medium and program product, which obtains a first data set and splits the first data set into multiple second data sets, wherein the multiple second data sets include: a training set and a verification set, wherein the training set includes: a supervised training set and a weakly supervised training set, wherein the supervised training set is a training set with accurate labels, and the weakly supervised training set is a training set without labels and / or with incorrect labels, and then determines the distribution parameters of the same feature in the multiple second data sets, and determines whether the feature is a feature to be deleted based on the multiple distribution parameters of each feature, wherein each feature corresponds to a distribution parameter in each second data set, and if the feature is a feature to be deleted, deletes the data corresponding to the feature from the second data set, and then performs supervised training on the first prediction model based on the supervised training set to obtain a second prediction model, and performs weakly supervised training on the second prediction model based on the training set to obtain a third prediction model, and finally verifies the performance of the third prediction model through the verification set. Compared with the prior art, due to the low quality of real data, the results obtained by machine learning and deep learning for predicting data are inaccurate and differ greatly from the actual results. The present application determines whether the characteristics of the data need to be deleted based on the distribution parameters in the data set. If the characteristics need to be deleted, the characteristics are deleted to improve the quality of the data set, so that subsequent model predictions are more accurate. And the first prediction model is supervised by the supervised training set to obtain the second prediction model, so that the model can learn the clear mapping relationship between the input data and the label, and then the second prediction model is weakly supervised according to the training set. The training set is then used to obtain the third prediction model. The model can further learn the potential structure and characteristics of the data, improve the robustness and generalization ability of the model, first conduct supervised training to ensure that the model obtains basic prediction capabilities; then weakly supervised training is performed to enable the model to adapt to more complex data scenarios, thereby further improving the prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0025] Figure 1 A schematic diagram of a scenario of a model training method provided in this application;
[0026] Figure 2 A flowchart of a first embodiment of a model training method provided in this application;
[0027] Figure 3 It is the characteristic distribution diagram of the training set and the validation set of the first data set;
[0028] Figure 4 is the feature box plot of the first data set;
[0029] Figure 5 is a linear relationship graph between the first data set and the final result;
[0030] Figure 6 Visualize the accuracy and fit of the trained second model in the supervised training set;
[0031] Figure 7 A flowchart of a second embodiment of a model training method provided in this application;
[0032] Figure 8 A flowchart of a third embodiment of a model training method provided in this application;
[0033] Fig. 9 A flowchart of a fourth embodiment of a model training method provided in this application;
[0034] Fig.10 A flowchart of a fifth embodiment of a model training method provided in this application;
[0035] Fig.11 A flowchart of a sixth embodiment of a model training method provided in this application;
[0036] Fig.12 A model training device provided in an embodiment of the present application;
[0037] Fig.13 An electronic device is provided in an embodiment of the present application.
[0038] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0039] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.
[0040] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.
[0041] In addition, this application involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and the use of artificial intelligence technology for automated decision-making, and a technical solution for making decisions that have a significant impact on personal rights and interests based on the results of automated decision-making. The application provides users with corresponding operation entrances for them to choose to agree or reject the results of automated decision-making; if the user chooses to reject, the expert decision-making process will be entered.
[0042] It should be noted that the model training method, device, equipment, storage medium and product provided in this application can be used in the field of big data, and can also be used in any field other than big data. The application field of the model training method, device, equipment, storage medium and product in this application is not limited.
[0043] In the existing technology, the dimensions and features of real data are manually labeled, there are a large number of repeated and useless features, and the data quality is not high. As a result, the results obtained by predicting data through machine learning and deep learning are quite different from the actual results, and the prediction results are inaccurate.
[0044] Based on this, the inventive concept of the present application is how to provide a model training method, device, equipment, storage medium and product, how to obtain high-quality data by processing the collected data, and then enable the model to process incomplete label information when training the model, so that the trained model can improve the robustness to noise data and further improve the prediction accuracy of the model. Therefore, the present application first pre-processes the data to be trained, deletes certain features that need to be deleted according to the distribution parameters of the data features, first improves the quality of the data, and then performs supervised training on the training set with accurate labels, so that the model can preliminarily learn the characteristics of the data, and then uses the training set with accurate labels and the training set without labels and / or incorrect labels to perform weakly supervised training on the model, so that the model can further learn the potential structure and characteristics of the data, improve the generalization ability of the model, and improve the robustness of the model to noise data, thereby improving the prediction accuracy of the data.
[0045] The present application provides a model training method, device, equipment, storage medium and program product, which are intended to solve the above technical problems in the prior art.
[0046] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.
[0047] Figure 1 A schematic diagram of a scenario of a model training method provided in this application, such as Figure 1 As shown, the server 101 obtains a first data set and splits the first data set into multiple second data sets, and then the server 101 determines the distribution parameters of the same feature in the multiple second data sets. The server 101 determines whether the feature is a feature to be deleted based on the multiple distribution parameters of each feature. If the feature is a feature to be deleted, the data corresponding to the feature is deleted from the second data set, wherein the server 101 can send the distribution parameters of the feature to the client 102 for display, and the result of determining whether the feature is a feature to be deleted can also be sent to the client 102 for display, so that the user can view it through the client 102. Finally, the server 101 performs supervised training on the first prediction model according to the supervised training set to obtain the second prediction model, and performs weakly supervised training on the second prediction model according to the training set to obtain the third prediction model, and verifies the performance of the third prediction model through the verification set, and the verification result is also sent to the client 102 for display, so that the user can view it.
[0048] Figure 2 A flow chart of a model training method embodiment 1 provided in this application, such as Figure 1 As shown, the method comprises the following steps:
[0049] S201: Obtain a first data set, and split the first data set into multiple second data sets, the multiple second data sets include: a training set and a validation set, the training set includes: a supervised training set and a weakly supervised training set, the supervised training set is a training set with accurate labels, and the weakly supervised training set is a training set without labels and / or with incorrect labels.
[0050] In the embodiment of the present application, a first data set is obtained, the first data set includes various historical data and information, and has multiple dimensional features. By analyzing the first data set and checking the feature distribution of the first data set, the first data set can be understood in advance and processed. For example, Figure 3 is the characteristic distribution diagram of the training set and validation set of the first data set, Figure 4 is the feature box plot of the first data set, Figure 5It is a linear relationship graph between the first data set and the final result.
[0051] The first data set is randomly and evenly split into multiple second data sets, each of which includes: a training set and a validation set, wherein the training set is further divided into a supervised training set and a weakly supervised training set, the supervised training set is a training set with accurate labels, and the weakly supervised training set is a training set without labels and / or with incorrect labels.
[0052] S202: Determine distribution parameters of the same feature in multiple second data sets, and determine whether the feature is a feature to be deleted based on multiple distribution parameters of each feature, where each feature corresponds to a distribution parameter in each second data set.
[0053] In an embodiment of the present application, distribution parameters of the same feature are calculated in each second data set, wherein each feature corresponds to a distribution parameter in each second data set, and whether the feature is a feature to be deleted is determined based on multiple distribution parameters of each feature.
[0054] S203: If the feature is a feature to be deleted, deleting the data corresponding to the feature from the second data set.
[0055] In the embodiment of the present application, if the feature is a feature to be deleted, the feature and the data corresponding to the feature are deleted from the second data set to obtain a clear data feature.
[0056] S204: Performing supervised training on the first prediction model according to the supervised training set to obtain a second prediction model, and performing weakly supervised training on the second prediction model according to the training set to obtain a third prediction model.
[0057] In an embodiment of the present application, supervised training is performed on the first prediction model according to supervised training sets in multiple second data sets to obtain a second prediction model, and weakly supervised training is performed on the second prediction model according to training sets in multiple second data sets to obtain a third prediction model.
[0058] Preferably, the first prediction model includes multiple base models. The first prediction model is adjusted through an objective function during supervised training. The output value of the objective function is positively correlated with the loss value and the total complexity of each base model. The loss value is used to indicate the degree of difference between the prediction value output by the first prediction model for the second data set and the label value of the second data set. The complexity of each base model is positively correlated with the number of nodes and edges included in the base model.
[0059] In an embodiment of the present application, the first prediction model includes multiple base models, and the multiple base models are trained one by one in sequence. Then, the prediction results of all the base models are linearly integrated to finally generate the final prediction result.
[0060]
[0061] in, is the prediction result of sample i after the tth iteration, is the prediction result of the first t-1 base models, f t (x i ) is the tth base model.
[0062] During the supervised training process, the first prediction model is adjusted by the objective function, and the output value of the objective function is positively correlated with the loss value and the total complexity of each base model:
[0063]
[0064] Where Obj represents the objective function L, represents the difference between the predicted value output by the first prediction model for the second data set and the label value of the second data set, Ω(f i ) represents the sum of the complexity of t base models, and the complexity of each base model is positively correlated to the number of nodes and edges included in the base model.
[0065] Figure 6 Visualize the accuracy and fit of the trained second model in the supervised training set. Figure 3 The leftmost column shows the correlation, which refers to the importance of the feature to the model's predictive ability. If the correlation graph shows that a feature is highly linearly correlated with the prediction target, then the feature has a significant contribution to the model's predictive ability. Figure 3 The middle part shows the standard deviation of the predicted value. The closer the standard deviation is to 0, the smaller the dispersion of the predicted value is. The prediction results of the model at each data point are more consistent, and the model fitting effect is better. Figure 3 The far right side shows the standardized prediction error. Data points with z distribution greater than 3 standard deviations are considered outliers or extreme cases. The z distribution graph shows that most data points are concentrated near z = 0, indicating that the prediction error of the model is small and the prediction is more accurate.
[0066] In an embodiment of the present application, when the first prediction model is supervised trained according to the supervised training set to obtain the second prediction model, the first prediction model is adjusted through the objective function, wherein the output value of the objective function is positively correlated with the loss value and the total complexity of each base model, the loss value measures the degree of fit of the model to the training data, and the total complexity of the base model is used as a regularization term to limit the complexity of the model and prevent the model from becoming too complex and losing its generalization ability. By balancing the loss value and complexity in the objective function, the problem of overfitting of the model can be prevented due to the model closely fitting the noise and outliers in the training data during the training process.
[0067] S205: Verify the performance of the third prediction model using a validation set.
[0068] In an embodiment of the present application, the performance of the third prediction model is verified by using a validation set from multiple second data sets. Optionally, the third prediction model is tested using the validation set to calculate indicators such as the model's accuracy and recall rate to determine whether the third prediction model meets the expected requirements and whether further adjustment or optimization is required.
[0069] In an embodiment of the present application, a first data set is obtained and split into multiple second data sets, the multiple second data sets include: a training set and a verification set, the training set includes: a supervised training set and a weakly supervised training set, the supervised training set is a training set with accurate labels, and the weakly supervised training set is a training set without labels and / or with incorrect labels, then the distribution parameters of the same feature in the multiple second data sets are determined, and based on the multiple distribution parameters of each feature, it is determined whether the feature is a feature to be deleted, each feature corresponds to a distribution parameter in each second data set, if the feature is a feature to be deleted, the data corresponding to the feature is deleted from the second data set, and then the first prediction model is supervised trained according to the supervised training set to obtain a second prediction model, and the second prediction model is weakly supervised trained according to the training set to obtain a third prediction model, and finally the performance of the third prediction model is verified through the verification set. Compared with the prior art, due to the low quality of real data, the results obtained by machine learning and deep learning for predicting data are inaccurate and differ greatly from the actual results. The present application determines whether the characteristics of the data need to be deleted based on the distribution parameters in the data set. If the characteristics need to be deleted, the characteristics are deleted to improve the quality of the data set, so that subsequent model predictions are more accurate. And the first prediction model is supervised by the supervised training set to obtain the second prediction model, so that the model can learn the clear mapping relationship between the input data and the label, and then the second prediction model is weakly supervised according to the training set. The training set is then used to obtain the third prediction model. The model can further learn the potential structure and characteristics of the data, improve the robustness and generalization ability of the model, first conduct supervised training to ensure that the model obtains basic prediction capabilities; then weakly supervised training is performed to enable the model to adapt to more complex data scenarios, thereby further improving the prediction accuracy of the model.
[0070] Figure 7 A flow chart of a second embodiment of a model training method provided in this application, such as Figure 7 As shown in the above Figure 2 Based on the embodiment shown, in step S204, a specific implementation method of performing weak supervision training on the second prediction model according to the training set to obtain the third prediction model is:
[0071] S701: For each feature, obtain one or more distribution sub-parameters of the feature in the training set.
[0072] In an embodiment of the present application, for each feature, one or more distribution sub-parameters of the feature in the training set are obtained, such as mean, variance, etc.
[0073] S702: For each data of each feature in the training set, generate noise data according to one or more distribution sub-parameters of the feature, and adjust the data of the feature by using the noise data to obtain a noisy data of the feature in the training set.
[0074] In the embodiment of the present application, for each data of each feature in the training set, noise data is generated according to one or more distribution sub-parameters of the feature. For example, Gaussian noise is generated according to the mean and variance of each feature. Gaussian noise is random noise that obeys Gaussian distribution. The generated noise data is added to the original feature data to obtain a noisy data of the feature in the training set.
[0075] S703: Perform weak supervision training on the second prediction model using the noisy data set consisting of the noisy data of each feature to obtain a third prediction model.
[0076] In an embodiment of the present application, weakly supervised training is performed on the second prediction model based on a noisy data set composed of noisy data of each feature. Due to the introduction of noise data, the training process will pay more attention to the model's capture of the overall trends and patterns of the data rather than the precise fitting of individual data points. Therefore, the third prediction model obtained has a higher tolerance to noise and uncertainty while maintaining a good fit to the original training data, thereby improving the generalization ability.
[0077] In an embodiment of the present application, when weakly supervised training is performed on the second prediction model according to the training set to obtain the third prediction model, for each feature, one or more distribution sub-parameters of the feature in the training set are obtained, and for each data of each feature in the training set, noise data is generated according to one or more distribution sub-parameters of the feature, and the data of the feature is adjusted by the noise data to obtain a noisy data of the feature in the training set, and the second prediction model is weakly supervised trained by the noisy data set composed of the noisy data of each feature to obtain the third prediction model. During the training process, the model learns to ignore small fluctuations and noise in the data, thereby improving the prediction ability for unseen data, and the model's adaptability to noise data is enhanced, which means that in actual applications, even if the model encounters inputs that are not completely consistent with the training data or contain noise, it can give relatively accurate predictions. Therefore, the robustness of the model is also enhanced by introducing noise data.
[0078] Preferably, the loss value in the weakly supervised training is obtained by weighting the first loss value of the supervised training set and the second loss value of the weakly supervised training set, and the weight of the first loss value is greater than the weight of the second loss value.
[0079] In an embodiment of the present application, in weakly supervised training, the loss value consists of two parts: the first loss value of the supervised training set and the second loss value of the weakly supervised training set. Therefore, the loss value in the weakly supervised training is obtained by weighting the first loss value of the supervised training set and the second loss value of the weakly supervised training set. Since the supervised training set provides accurate label information, the weight of the first loss value will be set larger to ensure that the model can fully learn this accurate information during the training process, and the weakly supervised training set has incomplete or uncertain label information, so the weight of the second loss value is set smaller to reduce the impact of this uncertain information on model training. Therefore, setting the weight of the first loss value greater than the weight of the second loss value can enable the model to fully learn accurate information, thereby improving the accuracy of the model, and when the supervised information is rich enough, the model can learn more generalized features and patterns, so that it can give more accurate prediction results when processing unseen data.
[0080] Figure 8 A flow chart of a model training method embodiment 3 provided in this application is shown in the above Figure 2 or Figure 7 Based on the embodiment shown, a specific implementation method of determining whether a feature is a feature to be deleted according to multiple distribution parameters of each feature in step S202 is:
[0081] S801: Variance of multiple distribution parameters of the feature.
[0082] In an embodiment of the present application, each feature corresponds to a distribution parameter in each second data set. For each feature, the distribution parameters in each second data set are extracted, and then the variance of these distribution parameters is calculated to obtain the variance of multiple distribution parameters of each feature.
[0083] S802: If the variance is greater than or equal to a preset variance threshold, the feature is determined to be a feature to be deleted.
[0084] In the embodiment of the present application, the variance reflects the degree of discreteness of the data set. If the variance of multiple distribution parameters of a feature is greater than or equal to a preset variance threshold, it indicates that the degree of discreteness of the multiple distribution parameters of the feature is too large, and the feature is determined to be a feature to be deleted.
[0085] In the embodiment of the present application, since each second data set is randomly screened, the same feature should be independent and identically distributed in each second data set. By calculating the variance of multiple distribution parameters of the feature, when the variance of the feature is greater than or equal to the preset variance threshold, that is, the feature is not independent and identically distributed in the second data set, it indicates that the feature is greatly affected by external factors. The feature may be caused by changes in market trends, etc., and it itself has little effect on the final result of the model, and may even have a negative impact on model training. Therefore, by deleting features with excessive variance, the generalization ability and prediction accuracy of the model can be improved.
[0086] Fig. 9 A flow chart of a fourth embodiment of a model training method provided in this application, in the above Figure 2 or Figure 7 Based on the illustrated embodiment, a specific implementation method of determining the distribution parameters of the same feature in multiple second data sets in step S202 is:
[0087] S901: Determine one or more distribution sub-parameters of the same feature in each second data set, where each feature corresponds to one or more distribution sub-parameters in each second data set.
[0088] In an embodiment of the present application, one or more distribution sub-parameters of the same feature in each second data set are calculated. For example, if the distribution sub-parameters are mean and variance, the mean and variance of the same feature in each second data set are calculated, that is, in each second data set, different features correspond to different means and variances.
[0089] S902: Perform comprehensive calculation on the distribution sub-parameters of each feature in each second data set to obtain the distribution parameters of the feature in each second data set.
[0090] In an embodiment of the present application, the distribution sub-parameters of each feature in each second data set are comprehensively calculated to serve as the distribution parameters of the feature in each second data set. For example, when the distribution sub-parameters are mean and variance, the distribution sub-parameters of each feature are combined into a composite index according to a weighted method, that is, the distribution parameters of the feature in each second data set are obtained, wherein the weight of each distribution sub-parameter is determined according to the actual situation and can be adjusted.
[0091] In an embodiment of the present application, when determining the distribution parameters of the same feature in multiple second data sets, the distribution sub-parameters of the same feature in multiple second data sets are first determined, and then the multiple distribution sub-parameters are comprehensively calculated to obtain a composite index, that is, the distribution parameters of the same feature in multiple second data sets. A unified distribution parameter is obtained by combining multiple distribution sub-parameters. While comprehensively considering multiple distribution sub-parameters, that is, mean, variance, etc., the distribution parameters of the feature are obtained by comprehensively calculating multiple distribution sub-parameters, which can make the subsequent step of determining whether the feature is a feature to be deleted simpler and more convenient, that is, the influence of all distribution sub-parameters on the feature can be comprehensively considered, and the calculation process of determining the feature to be deleted can be simplified.
[0092] Fig.10 A flow chart of a fifth embodiment of a model training method provided in this application is shown in the above Fig. 9 Based on the embodiment shown, before step S901, the method further includes:
[0093] S1001: For each feature, obtain target data of the feature in the first data set, and calculate the average value of the target data.
[0094] In the embodiment of the present application, for each feature, the target data of the feature in the first data set is obtained, and then the average value of the target data is calculated.
[0095] S1002: Update the target data whose difference with the average value is greater than or equal to the preset difference value to the average value.
[0096] In an embodiment of the present application, for each feature, the target data is traversed and the difference between the target data and the average value is calculated. If the difference between the target data and the average value is greater than or equal to the preset difference, it indicates that the target data is an outlier data under the feature, and the target data whose difference with the average value is greater than or equal to the preset difference is updated to the average value.
[0097] In the embodiments of the present application, the reason for the appearance of outliers under each feature may come from market fluctuations, emergencies or the issuance of relevant documents, etc., and these outliers have little correlation with the prediction target of the model. Therefore, in order to avoid the negative impact of outliers on the model, each feature is processed and the average value of the target data under each feature is used to replace the outlier under each feature, so as to reduce the impact on model training and improve the accuracy of model prediction.
[0098] Fig.11 A flow chart of a sixth embodiment of a model training method provided in this application, in the above Figure 2 or Figures 7 to 10 Based on any of the embodiments shown, before step S204, the method further includes:
[0099] S1101: Determine the total difference degree of distribution parameters of each feature in multiple second data sets.
[0100] In an embodiment of the present application, before performing supervised training on the first prediction model according to a supervised training set to obtain a second prediction model, and performing weakly supervised training on the second prediction model according to the training set to obtain a third prediction model, it is necessary to determine the total degree of difference in the distribution parameters of each feature in multiple second data sets. Optionally, after obtaining the distribution parameters of each feature in multiple second data sets, for each feature, the absolute difference in its distribution parameters between different second data sets is calculated, and the sum of the absolute differences in the distribution parameters between different second data sets is taken as the degree of difference in the distribution parameters of the feature in the second data sets.
[0101] S1102: If the total difference degree of the distribution parameters is less than or equal to the preset threshold, the step of performing supervised training on the first prediction model according to the supervised training set to obtain the second prediction model, and performing weakly supervised training on the second prediction model according to the training set to obtain the third prediction model.
[0102] In an embodiment of the present application, if the total difference degree of the distribution parameters of each feature is less than or equal to a preset threshold, the first prediction model is supervised trained according to the supervised training sets in multiple second data sets to obtain the second prediction model, and the second prediction model is weakly supervised trained according to the training sets in multiple second data sets to obtain the third prediction model.
[0103] In an embodiment of the present application, before performing supervised training on the first prediction model according to the supervised training set to obtain the second prediction model, and performing weakly supervised training on the second prediction model according to the training set to obtain the third prediction model, the total difference degree of distribution parameters of each feature in multiple second data sets is determined. If the total difference degree of the distribution parameters is less than or equal to the preset threshold, the step of performing supervised training on the first prediction model according to the supervised training set to obtain the second prediction model, and performing weakly supervised training on the second prediction model according to the training set to obtain the third prediction model is entered. When the total difference degree of the distribution parameters is less than or equal to the preset threshold, it means that the training set and the validation set are the same in feature distribution. Checking the consistency of data distribution before training helps to ensure that the model can learn the essential characteristics of the data and improve the generalization ability of the model.
[0104] Fig.12 A model training device provided in an embodiment of the present application includes: an acquisition module 1201 and a processing module 1202; wherein,
[0105] The acquisition module 1201 is used to acquire a first data set and split the first data set into multiple second data sets, wherein the multiple second data sets include: a training set and a verification set, wherein the training set includes: a supervised training set and a weakly supervised training set, wherein the supervised training set is a training set with accurate labels, and the weakly supervised training set is a training set with no labels and / or incorrect labels; the processing module 1202 is used to determine the distribution parameters of the same feature in multiple second data sets, and determine whether the feature is a feature to be deleted based on the multiple distribution parameters of each feature, wherein each feature corresponds to a distribution parameter in each second data set; the processing module 1202 is also used to delete the data corresponding to the feature from the second data set when the feature is a feature to be deleted; the processing module 1202 is also used to perform supervised training on the first prediction model according to the supervised training set to obtain a second prediction model, and perform weakly supervised training on the second prediction model according to the training set to obtain a third prediction model; the processing module 1202 is also used to verify the performance of the third prediction model through the verification set.
[0106] A model training device provided in the present application also includes: a processing module 1202, which is used to adjust the first prediction model through an objective function during a supervised training process.
[0107] A model training device provided by the present application also includes: for each feature, an acquisition module 1201 is used to obtain one or more distribution sub-parameters of the feature in the training set; for each data of each feature in the training set, a processing module 1202 is used to generate noise data according to one or more distribution sub-parameters of the feature, and adjust the data of the feature through the noise data to obtain a noisy data of the feature in the training set; the processing module 1202 is also used to perform weakly supervised training on the second prediction model through a noisy data set composed of the noisy data of each feature to obtain a third prediction model.
[0108] A model training device provided in the present application also includes: based on the variance of multiple distribution parameters of the feature, if the variance is greater than or equal to a preset variance threshold, a processing module 1202 is used to determine the feature as a feature to be deleted.
[0109] A model training device provided in the present application also includes: a processing module 1202, which is used to determine one or more distribution sub-parameters of the same feature in each second data set, and each feature corresponds to one or more distribution sub-parameters in each second data set; the processing module 1202 is also used to perform comprehensive calculations on the distribution sub-parameters of each feature in each second data set as the distribution parameters of the feature in each second data set.
[0110] A model training device provided in the present application also includes: for each feature, an acquisition module 1201, used to obtain target data of the feature in the first data set, and a processing module 1202, used to calculate the average value of the target data; the processing module 1202 is also used to update the target data whose difference with the average value is greater than or equal to a preset difference to the average value.
[0111] A model training device provided in the present application also includes: a processing module 1202, which is used to determine the total difference degree of distribution parameters of each feature in multiple second data sets; if the total difference degree of the distribution parameters is less than or equal to a preset threshold, then enter the step of performing supervised training on the first prediction model according to the supervised training set to obtain a second prediction model, and performing weakly supervised training on the second prediction model according to the training set to obtain a third prediction model.
[0112] Fig.13 An electronic device provided in an embodiment of the present application includes: at least one processor 1301 and a memory 1302 . The processor 1301 and the memory 1302 are connected via a bus 1303 .
[0113] The specific implementation process of the processor 1301 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.
[0114] In the above Fig.13 In the illustrated embodiment, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.
[0115] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (Non-volatile Memory, NVM), such as at least one disk storage.
[0116] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.
[0117] The present application also provides a computer-readable storage medium on which computer execution instructions are stored. When a processor executes the computer execution instructions, a model training method as described above is implemented.
[0118] The computer-readable storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The computer-readable storage medium may be any available medium that can be accessed by a general or special purpose computer.
[0119] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.
[0120] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.
[0121] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0122] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
[0123] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0124] The present application also provides a computer program product, including a computer program, which, when executed by a processor, is used to implement a model training method as described above.
[0125] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.
[0126] It should be further noted that, although the various steps in the flowchart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
[0127] It should be understood that the above-mentioned device embodiments are only illustrative, and the device of the present application can also be implemented in other ways. For example, the division of units / modules in the above-mentioned embodiments is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units, modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed.
[0128] In addition, unless otherwise specified, each functional unit / module in each embodiment of the present application may be integrated into one unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated together. The above-mentioned integrated unit / module may be implemented in the form of hardware or in the form of a software program module.
[0129] If the integrated unit / module is implemented in the form of hardware, the hardware may be a digital circuit, an analog circuit, etc. The physical implementation of the hardware structure includes but is not limited to transistors, memristors, etc. If not specifically stated, the processor may be any appropriate hardware processor, such as a CPU, a GPU, an FPGA, a DSP, an ASIC, etc. If not specifically stated, the storage unit may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc.
[0130] If the integrated unit / module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.
[0131] In the above embodiments, the description of each embodiment has its own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0132] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.
[0133] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.
Claims
1. A model training method, characterized in that: include: Acquire a first data set, and split the first data set into multiple second data sets, the multiple second data sets include: a training set and a validation set, the training set includes: a supervised training set and a weakly supervised training set, the supervised training set is a training set with accurate labels, and the weakly supervised training set is a training set without labels and / or with incorrect labels; Determine distribution parameters of the same feature in multiple second data sets, and determine whether the feature is a feature to be deleted according to the multiple distribution parameters of each feature, each feature corresponds to one distribution parameter in each second data set; If the feature is a feature to be deleted, deleting the data corresponding to the feature from the second data set; Performing supervised training on the first prediction model according to the supervised training set to obtain a second prediction model, and performing weakly supervised training on the second prediction model according to the training set to obtain a third prediction model; The performance of the third prediction model is verified using the verification set.
2. The method according to claim 1, characterized in that The first prediction model includes multiple base models. During the supervised training process, the first prediction model is adjusted through an objective function. The output value of the objective function is positively correlated with the loss value and the total complexity of each base model. The loss value is used to indicate the degree of difference between the prediction value output by the first prediction model for the second data set and the label value of the second data set. The complexity of each base model is positively correlated with the number of nodes and the number of edges included in the base model.
3. The method according to claim 1, characterized in that The step of performing weak supervision training on the second prediction model according to the training set to obtain a third prediction model includes: For each feature, obtaining one or more distribution sub-parameters of the feature in the training set; For each data of each feature in the training set, generate noise data according to the one or more distribution sub-parameters of the feature, and adjust the data of the feature by using the noise data to obtain a noisy data of the feature in the training set; The second prediction model is subjected to weak supervision training using a noisy data set consisting of the noisy data of each of the features to obtain a third prediction model.
4. The method according to claim 3, characterized in that The loss value in the weakly supervised training is obtained by weighting the first loss value of the supervised training set and the second loss value of the weakly supervised training set, and the weight of the first loss value is greater than the weight of the second loss value.
5. The method according to any one of claims 1 to 4, characterized in that The determining whether the feature is a feature to be deleted according to the plurality of distribution parameters of each feature comprises: a variance of a plurality of said distribution parameters according to said features; If the variance is greater than or equal to a preset variance threshold, the feature is determined to be a feature to be deleted.
6. The method according to any one of claims 1 to 4, characterized in that The determining of the distribution parameters of the same feature in the plurality of the second data sets comprises: Determine one or more distribution sub-parameters of the same feature in each of the second data sets, each of the features corresponding to one or more distribution sub-parameters in each of the second data sets; The distribution sub-parameters of each of the features in each of the second data sets are comprehensively calculated as the distribution parameters of the features in each of the second data sets.
7. The method according to claim 6, characterized in that Before determining the distribution parameters of the same feature in the plurality of the second data sets, the method further comprises: For each of the features, obtaining target data of the feature in the first data set, and calculating an average value of the target data; The target data whose difference with the average value is greater than or equal to a preset difference is updated to the average value.
8. The method according to claim 6, characterized in that Before the supervised training of the first prediction model according to the supervised training set to obtain the second prediction model and the weakly supervised training of the second prediction model according to the training set to obtain the third prediction model is performed, the method further includes: Determining the total degree of difference of the distribution parameters of each of the features in a plurality of the second data sets; If the total degree of difference of the distribution parameters is less than or equal to a preset threshold, the step of performing supervised training on the first prediction model according to the supervised training set to obtain a second prediction model, and performing weakly supervised training on the second prediction model according to the training set to obtain a third prediction model is entered.
9. A model training device, characterized in that: include: An acquisition module is used to acquire a first data set and split the first data set into multiple second data sets, wherein the multiple second data sets include: a training set and a validation set, wherein the training set includes: a supervised training set and a weakly supervised training set, wherein the supervised training set is a training set with accurate labels, and the weakly supervised training set is a training set with no labels and / or incorrect labels; A processing module, used to determine distribution parameters of the same feature in multiple second data sets, and determine whether the feature is a feature to be deleted according to the multiple distribution parameters of each feature, each feature corresponding to one distribution parameter in each second data set; The processing module is further configured to delete data corresponding to the feature from the second data set when the feature is a feature to be deleted; The processing module is further used to perform supervised training on the first prediction model according to the supervised training set to obtain a second prediction model, and to perform weakly supervised training on the second prediction model according to the training set to obtain a third prediction model; The processing module is further used to verify the performance of the third prediction model through the verification set.
10. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 8.
11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 8 when executed by a processor.
12. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 8 when being executed by a processor.