WO2021139465A1 - Backward model selection method and device, and readable storage medium - Google Patents

Abstract

A backward model selection method and device, and a readable storage medium. The backward model selection method comprises: receiving configuration parameters sent by a client associated with a server and acquiring features to be trained, and training, on the basis of said features and the configuration parameters, a preset model to be trained to obtain a first initial training model (S10); calculating first significances corresponding to said features, and removing, on the basis of the first significances, features to be removed satisfying a preset removal significance requirement from the features to be trained, so as to perform loop training on the first initial training model on the basis of the removed features to be trained to obtain a loop training model set (S20); selecting a target training model from the first initial training model and the loop training model set on the basis of the configuration parameters (S30); and generating visual data corresponding to the target training model, and feeding back the visual data to the client (S40).

Description

Backward model selection method, equipment and readable storage medium

Cross-references to related applications

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on January 9, 2020, the application number is 202010024439.3, and the application name is "backward model selection method, equipment and readable storage medium", the entire content of which is incorporated by reference Incorporated in this application.

Technical field

This application relates to the artificial intelligence technology field of Fintech, and in particular to a backward model selection method, device and readable storage medium.

Background technique

With the continuous development of financial technology, especially Internet technology and finance, more and more technologies (such as distributed, blockchain, artificial intelligence, etc.) are applied in the financial field, but the financial industry has also proposed higher technology Requirements, such as the distribution of to-do items corresponding to the financial industry, also have higher requirements.

With the continuous development of computer software and artificial intelligence, the application of machine learning modeling has become more and more extensive. In the existing technology, financial risk control, medical models and other scenarios usually use logistic regression model modeling, and logistic regression In model modeling, the backward selection mode is an important model selection strategy. Compared with all the features added to the model training, it can effectively prevent the model from overfitting. However, the current backward selection mode usually requires the modeler to have High code development capabilities, and can only be implemented in a single machine, that is, the current implementation of the backward selection mode has higher threshold requirements for modelers, and because it can only be implemented in a single machine, it leads to the backward selection mode. The modeling time is long and the modeling efficiency is low. Therefore, the prior art has the technical problems of high modeling threshold and low efficiency of the backward selection mode.

Application content

The main purpose of this application is to provide a backward model selection method, device and readable storage medium, aiming to solve the technical problems of high modeling threshold and low efficiency of backward selection mode in the prior art.

To achieve the above objective, the present application provides a backward model selection method, the backward model selection method is applied to the server, and the backward model selection method includes:

Receiving configuration parameters sent by the client associated with the server and acquiring features to be trained, and training a preset model to be trained based on each of the features to be trained and the configuration parameters to obtain a first initial training model;

Calculate the first saliency corresponding to each of the features to be trained, and based on each of the first saliency, eliminate the features to be removed that meet the preset saliency requirements for removal from the features to be trained, so as to be based on the removed features. Each of the features to be trained performs cyclic training on the first initial training model to obtain a cyclic training model set;

Based on the configuration parameters, selecting a target training model from the first initial training model and the cyclic training model set;

Generate visualization data corresponding to the target training model, and feed back the visualization data to the client.

In an embodiment, the cyclic training model set includes one or more model elements, and each of the model elements includes a second initial training model,

According to each of the first saliency, the feature to be removed that meets the preset removal saliency requirement is removed from the features to be trained, so as to compare the first initial saliency based on the removed features to be trained. The training model performs cyclic training, and the steps to obtain the cyclic training model set include:

Based on each of the first saliency and the preset removal saliency requirement, select the feature to be removed from the features to be trained, and remove the feature to be removed;

Training the first initial training model based on each of the features to be trained after being eliminated to obtain the second initial training model;

Calculate the second saliency of each feature to be trained after culling, and based on each of the second saliency, remove other features that meet the preset removal saliency requirements from the features to be trained after removal. The features to be removed;

Based on each of the features to be trained after being removed again, the second initial training model is cyclically trained to obtain one or more of the model elements until the feature to be removed does not exist in each of the features to be trained.

In an embodiment, the step of selecting the feature to be removed from the features to be trained based on each of the first saliency and the preset removal saliency requirement includes:

Comparing each of the first saliency, and selecting the feature with the lowest saliency among the features to be trained as the target feature;

Comparing the target saliency of the target feature with a preset saliency rejection threshold;

If the target significance is less than the preset rejection significance threshold, it is determined that the target feature meets the preset rejection significance requirement, and the target feature is taken as the feature to be rejected.

In an embodiment, the step of calculating the first saliency corresponding to each of the features to be trained includes:

Calculating the chi-square value wald of each of the features to be trained;

Based on each of the chi-square value wald and the degrees of freedom of each of the features to be trained, each of the first saliences is calculated.

In an embodiment, the configuration parameter includes a training completion determination condition, and the feature to be trained includes one or more pieces of feature data;

The step of training a preset model to be trained based on each of the features to be trained and the configuration parameters, and obtaining a first initial training model includes:

Input the feature data corresponding to each of the features to be trained into the preset model to be trained, so as to train and update the preset model to be trained;

Judging whether the updated preset to-be-trained model satisfies the training completion judgment condition, and if the updated preset to-be-trained model satisfies the training completion judgment condition, the first initial training model is obtained;

If the updated preset to-be-trained model does not meet the training completion judgment condition, continue to perform iterative training updates on the preset to-be-trained model until the updated preset to-train model satisfies the training Complete the judgment condition.

In an embodiment, the step of selecting a target training model from the first initial training model and the cyclic training model set based on the configuration parameters includes:

Acquire the model selection strategy in the parameter configuration, where the model selection strategy includes AUC (Area Under Curve, the area under the receiver operating characteristic curve and the coordinate axis) value and AIC (Akaike information criterion, Akaike information) Quantity criterion) value;

If the model selection strategy is the AUC value, compare the AUC values of the elements in the cyclic training model set, and select the element corresponding to the largest AUC value as the target training model;

If the model selection strategy is the AIC value, the AIC values of the elements in the cyclic training model set are compared, and the element corresponding to the smallest AIC value is selected as the target training model.

In an embodiment, the client includes a visual interface,

The step of generating visualization data corresponding to the target training model and feeding back the visualization data to the client includes:

Acquiring candidate feature data, selection summary data, and training process data corresponding to the backward model selection process of the target training model;

Generate visualization data corresponding to the candidate feature data, the selection summary data, and the training process data, and feed back the visualization data to the visualization interface in real time.

To achieve the above objective, the present application also provides a backward model selection method. The backward model selection method is applied to the client, and the backward model selection method includes:

Receive a model selection task, and send the configuration parameters corresponding to the model selection task to the server associated with the client, so that the server performs model selection based on the configuration parameters and the acquired features to be trained to obtain A target training model, and obtaining visualization data corresponding to the target training model, so as to send the visualization data to the client;

The visualization data fed back by the server is received, and the visualization data is displayed on a preset visualization interface.

The present application also provides a backward model selection device, which is applied to a backward model selection device, and the backward model selection device includes:

The first training module is configured to receive the configuration parameters sent by the client associated with the server and obtain the features to be trained, and train a preset model to be trained based on each of the features to be trained and the configuration parameters , To obtain the first initial training model;

The second training module is used to calculate the first saliency corresponding to each of the features to be trained, and based on each of the first saliences, to remove the features that meet the preset removal saliency requirements from the features to be trained The features to be eliminated, to perform cyclic training on the first initial training model based on each of the features to be trained after culling, to obtain a cyclic training model set;

A selection module for selecting a target training model from the first initial training model and a set of cyclic training models based on the configuration parameters;

The feedback module is used for generating the visualization data corresponding to the target training model, and feeding back the visualization data to the client.

In an embodiment, the second training module includes:

The first culling sub-module is configured to select the feature to be removed among the features to be trained based on each of the first saliency and the preset saliency removal requirement, and to remove the feature to be removed ；

A training sub-module, configured to train the first initial training model based on the eliminated features to be trained to obtain the second initial training model;

The second culling sub-module is used to calculate the second saliency of each feature to be trained after being removed, and based on each of the second saliency, remove the coincidence again from each feature to be trained after being removed Other features to be removed that are required to be removed by the preset saliency;

The cyclic training sub-module is used to perform cyclic training on the second initial training model based on each of the features to be trained after being removed again, to obtain one or more of the model elements, until each feature to be trained The feature to be removed does not exist in.

In an embodiment, the selection submodule includes:

The first comparison unit is configured to compare each of the first saliency, and select the feature with the lowest saliency among the features to be trained as the target feature;

The second comparison unit is used to compare the target significance of the target feature with a preset significance threshold for rejection;

The determining unit is configured to determine that if the target significance is less than the preset rejection significance threshold, determine that the target feature meets the preset rejection significance requirement, and use the target feature as the pending Remove features.

In an embodiment, the second training module further includes:

The first calculation sub-module is used to calculate the chi-square value wald of each of the features to be trained;

The second calculation sub-module is used for calculating each of the first saliency based on each of the chi-square value wald and the degrees of freedom of each of the features to be trained.

In an embodiment, the first training module includes:

A training update sub-module for inputting the feature data corresponding to each of the features to be trained into the preset model to be trained, so as to train and update the preset model to be trained;

The first judging sub-module is used to judge whether the updated preset model to be trained satisfies the training completion judging condition, and if the updated preset to be trained model satisfies the training completion judging condition, then Obtaining the first initial training model;

The second judgment sub-module is configured to continue to perform iterative training updates on the preset to-be-trained model if the updated preset to-be-trained model does not satisfy the training completion judgment condition until the updated all-in-one model The preset model to be trained satisfies the training completion judgment condition.

In an embodiment, the selection module includes:

The first obtaining sub-module is configured to obtain the model selection strategy in the parameter configuration, wherein the model selection strategy includes an AUC value and an AIC value;

The first comparison sub-module is configured to compare the AUC value of each element in the cyclic training model set if the model selection strategy is the AUC value to select the largest corresponding AUC value As the target training model;

The second comparison sub-module is used to compare the AIC value of each element in the cyclic training model set if the model selection strategy is the AIC value to select the smallest corresponding AIC value As the target training model.

In an embodiment, the feedback module includes:

The second acquisition sub-module is used to acquire the candidate feature data, selection summary data, and training process data corresponding to the backward model selection process of the target training model;

A generating sub-module is used to generate the visualization data corresponding to the candidate feature data, the selection summary data, and the training process data, and feed back the visualization data to the visualization interface in real time.

In order to achieve the above objective, the present application also provides a backward model selection device. The backward model selection device is applied to a client, and the backward selection device includes:

The sending module is configured to receive the model selection task, and send the configuration parameters corresponding to the model selection task to the server associated with the client, so that the server can use the configuration parameters and the acquired configuration parameters. Performing model selection on training features, obtaining a target training model, and obtaining visualization data corresponding to the target training model, so as to send the visualization data to the client;

The receiving module is configured to receive the visualization data fed back by the server, and display the visualization data on a preset visualization interface.

The present application also provides a backward model selection device. The backward model selection device includes a memory, a processor, and a device for the backward model selection method that is stored on the memory and can run on the processor. A program, when the program of the backward model selection method is executed by a processor, the steps of the backward model selection method as described above can be realized.

The present application also provides a readable storage medium, the readable storage medium stores a program for implementing the backward model selection method, and when the program for the backward model selection method is executed by a processor, the backward model as described above is implemented Select the steps of the method.

This application receives the configuration parameters sent by the client associated with the server and obtains the features to be trained, and trains a preset model to be trained based on each of the features to be trained and the configuration parameters to obtain the first initial training The model further calculates the first saliency corresponding to each of the features to be trained, and based on each of the first salience, removes the features to be removed that meet the preset removal saliency requirements from the features to be trained, and then based on After removing each of the features to be trained, the first initial training model is cyclically trained to obtain a cyclic training model set, and then based on the configuration parameters, from the first initial training model and the cyclic training model set The target training model is selected, and then the visualization data corresponding to the target training model is generated, and the visualization data is fed back to the client. That is, this application first sends the configuration parameters sent by the client associated with the server and acquires the features to be trained, and based on each of the features to be trained and the configuration parameters, performs a comparison of the preset model to be trained Training, obtain the first initial training model, and then perform the calculation of the first saliency corresponding to each of the features to be trained, and then based on each of the first saliency, the elimination of the features to be trained meets the preset elimination The features to be removed with the saliency requirement are further based on the removed features to be trained, the cyclic training of the first initial training model is performed to obtain the cyclic training model set, and then based on the configuration parameters, from the first A target training model is selected from an initial training model and a cyclic training model set, and then the visualization data corresponding to the target training model is generated, and the visualization data is fed back to the client. That is, this application provides a model selection method of backward selection mode of codeless distributed modeling and visual modeling. The user only needs to set and send the necessary configuration parameters to the server through the client, and the server is It can feed back the visual data and the result of the backward model selection process corresponding to the corresponding backward model selection process, that is, through the communication connection between the client and the server for model modeling, distributed modeling is realized, and compared with a single machine The modeling of the backward selection mode performed improves the modeling efficiency of the backward selection mode. By generating the visualization data corresponding to the target training model and feeding it back to the client, the visualization modeling is realized and the construction is reduced. The ability threshold of the model personnel is required and the modeling efficiency of the backward selection mode is further improved. In this application, the user only needs to enter the necessary model parameters in the visual interface of the client to obtain the corresponding backward model selection results. There is no requirement for code development ability, which realizes no-code modeling, and further reduces the requirement for the ability threshold of modelers. Therefore, it solves the technology of high modeling threshold and low efficiency of backward selection mode in the existing technology. problem.

Description of the drawings

The drawings herein are incorporated into the specification and constitute a part of the specification, show embodiments that conform to the application, and are used together with the specification to explain the principle of the application.

In order to more clearly describe the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, those of ordinary skill in the art are In other words, other drawings can be obtained based on these drawings without creative labor.

FIG. 1 is a schematic flowchart of the first embodiment of the backward model selection method of this application;

2 is a schematic diagram of a visual interface for configuring the parameters in the backward model selection method of this application;

3 is a schematic flowchart of a second embodiment of the backward model selection method of this application;

4 is a schematic diagram of the process of performing backward model selection in combination with the first embodiment in the second embodiment of the backward model selection method of this application;

FIG. 5 is a schematic flowchart of a third embodiment of a backward model selection method according to this application;

FIG. 6 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the application.

The realization, functional characteristics, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

The embodiment of the present application provides a method for selecting a backward model. The method for selecting a backward model is applied to the server. In the first embodiment of the method for selecting a backward model of the present application, referring to FIG. 1, the backward model selection is Methods include:

Step S10, receiving configuration parameters sent by the client associated with the server and acquiring features to be trained, and training a preset model to be trained based on each of the features to be trained and the configuration parameters to obtain a first initial training model.

In this embodiment, it should be noted that the client includes a visualization interface, and the user can configure parameters of a preset model to be trained on the visualization interface for model training, as shown in FIG. In the visual interface of parameter configuration, the parameters such as the maximum iteration coefficient, minimum convergence error, and category weight are all parameters that need to be set before model training. The backward model selection mode includes backward selection mode and stepwise selection mode. The feature to be trained includes one or more features, and each feature includes one to obtain multiple pieces of feature data. The preset model to be trained includes a logistic regression model.

The configuration parameters sent by the client associated with the server are received and the features to be trained are acquired, and a preset model to be trained is trained based on each of the features to be trained and the configuration parameters to obtain a first initial training model. Specifically, the configuration parameters sent by the client are received, and training completion judgment conditions are extracted from the configuration parameters, and then each feature to be trained is obtained from the local database of the backward model selection server, and each feature to be trained is The feature data corresponding to the feature is input into the preset to-be-trained model to perform iterative training updates on the preset to-be-trained model, until the preset to-be-trained model reaches the preset training completion judgment condition, then the iterative training is completed, and The updated preset model to be trained, that is, the first initial training model is obtained, wherein the preset training completion judgment condition includes reaching the minimum convergence error, reaching the maximum number of iterations, and so on.

Wherein, the configuration parameters include training completion judgment conditions, and the features to be trained include one or more pieces of feature data;

The step of training a preset model to be trained based on each of the features to be trained and the configuration parameters, and obtaining a first initial training model includes:

Step S11, input the feature data corresponding to each of the features to be trained into the preset model to be trained, so as to train and update the preset model to be trained;

In this embodiment, it should be noted that each time the preset model to be trained is trained once, the preset model to be trained is updated once, wherein the preset model to be trained is trained and updated The gradient descent method and so on.

Step S12: Determine whether the updated preset to-be-trained model satisfies the training completion judgment condition, and if the updated preset to-be-trained model satisfies the training completion judgment condition, obtain the first initial training model;

In this embodiment, it should be noted that the training completion judgment condition includes reaching the minimum convergence error, reaching the maximum number of iterations, and so on.

It is judged whether the updated preset to-be-trained model satisfies the training completion judgment condition, and if the updated preset to-be-trained model satisfies the training completion judgment condition, the first initial training model is obtained, specifically To determine whether the updated preset to-be-trained model satisfies the training completion judgment condition, and if the updated preset to-be-trained model satisfies the training completion judgment condition, the updated model obtained in this training The preset model to be trained is used as the first initial training model, that is, the first initial training model is obtained.

Step S13: If the updated preset to-be-trained model does not meet the training completion judgment condition, continue to perform iterative training updates on the preset to-be-trained model until the updated preset to-be-trained model satisfies The training completion judgment condition.

In this embodiment, if the updated preset to-be-trained model does not meet the training completion determination condition, then iterative training and update of the preset to-be-trained model continues until the updated preset to-be-trained model The training model satisfies the training completion judgment condition. Specifically, if the updated preset model to be trained does not satisfy the training completion judgment condition, it indicates that the updated preset model to be trained obtained in this training Cannot be used as the first initial training model, and then input the feature data corresponding to each of the features to be trained into the updated preset model to be trained, so as to perform iterative training updates on the preset model to be trained, Until the updated preset to-be-trained model satisfies the training completion judgment condition.

Step S20: Calculate the first saliency corresponding to each of the features to be trained, and based on each of the first saliency, remove the features to be removed that meet the preset saliency removal requirements from the features to be trained, so as to be based on After removing each of the features to be trained, performing cyclic training on the first initial training model to obtain a cyclic training model set;

In this embodiment, the first saliency corresponding to each of the features to be trained is calculated, and based on each of the first saliency, the features to be removed that meet the preset removal saliency requirements are eliminated from the features to be trained , To perform cyclic training on the first initial training model based on the removed features to be trained to obtain a cyclic training model set, specifically, based on each of the features to be trained and the features corresponding to each of the features to be trained As a result of model training, the chi-square value wald of each feature to be trained is calculated by the preset chi-square value wald calculation formula, and then based on each chi-square value wald and the degrees of freedom of each feature to be trained, the corresponding to each feature to be trained is calculated The first saliency of, and then based on each of the first saliency, find and remove the feature to be removed in each of the features to be trained, and then based on the feature to be trained after removing the feature to be removed, the The first initial training model is re-trained and updated, and the updated first initial training model is obtained, that is, one of the model elements of the cyclic training model set is obtained, and further, in each of the features to be trained after removal The search for the features to be eliminated and the training of the updated first initial training model are performed again to obtain model elements until each feature to be trained does not have the features to be eliminated. At this time, one or more The model element, that is, the cyclic training model set is obtained.

Wherein, in step S20, the step of calculating the first saliency corresponding to each of the features to be trained includes:

Step S21: Calculate the chi-square value wald of each of the features to be trained;

In this embodiment, the chi-square value wald of each feature to be trained is calculated, specifically, the feature data representation matrix corresponding to each feature to be trained is substituted into the preset chi-square value wald calculation formula, and each of the features is calculated in parallel. The chi-square value wald corresponding to the feature to be trained, wherein the preset chi-square value wald calculation formula is as follows:

among them,

Where S is the chi-square value wald, and the feature data corresponding to the feature to be trained is denoted as X, where X includes n pieces of data, each piece of data includes k values, and X can be represented by a feature data representation matrix. The feature data indicates that each column of the matrix is a piece of data and corresponds to the feature to be trained, and the model parameter obtained by training the preset model to be trained corresponding to X is θ, where θ is a k-dimensional vector (θ ₁ , Θ ₂ , ..., θ _k-1 , θ _k ), and the feature set X to be trained can be divided into a first model feature set and a second model feature set, wherein the feature corresponding to the first model feature set The data representation matrix is X0, the feature data representation matrix corresponding to the second model feature set is X1, X ₀ includes n pieces of data, each piece of data includes (kt) values, and X ₀ trains the preset model to be trained The model parameter obtained is θ ₀ , where θ ₀ is a (kt)-dimensional vector (θ ₁ , θ ₂ ,..., θ _kt ), X ₁ includes n pieces of data, and each piece of data includes t values. The data set corresponding to the target output of the training model is Y, where Y includes n pieces of data, and Y corresponds to the predicted probability P, and P includes n probabilities (p ₁ , p ₂ , ..., p _n-1 , p _n ) , Then the null hypothesis H ₀ : Cθ=h is performed at this time, at this time all values are 0, C is a matrix of t*k, and h is a vector of k*1. Further, based on each of the chi-square values wald, Eliminate the non-saliency features in the features to be trained to obtain the second feature to be trained, where the non-saliency features refer to the features of the features to be trained that are significantly less than a preset significance threshold , Wherein the saliency can be obtained based on the chi-square value wald and the degree of freedom of the feature to be trained, wherein the degree of freedom is related to the value of the feature to be trained, for example, suppose the feature to be trained Including bank deposits, credit card consumption records, and loan records, then the feature to be trained includes 3 variables, and the degree of freedom is 2.

Step S22: Calculate each of the first saliency based on each of the chi-square value wald and the degrees of freedom of each of the features to be trained.

In this embodiment, it should be noted that the first significance can be determined based on the Pearson correlation value, and when the Pearson correlation value is less than or equal to the preset Pearson correlation threshold, the determination is The feature corresponding to the first saliency does not meet the preset saliency removal requirement, that is, the feature corresponding to the first saliency appears to be significant, when the Pearson correlation value is greater than the preset Pearson correlation threshold When, it is determined that the feature corresponding to the first saliency satisfies the preset saliency removal requirement, that is, the feature corresponding to the first saliency appears to be insignificant, and the degree of freedom corresponds to the number of feature data corresponding to the feature Correlation, for example, assuming that there are 100 different pieces of data in the feature data, the degree of freedom is 99.

Calculate each of the first saliency based on each of the chi-square value wald and the degrees of freedom of each of the features to be trained, specifically, based on each of the chi-square value wald and the degrees of freedom of each feature to be trained, The Pearson correlation value of each feature to be trained is calculated by a preset Pearson correlation value calculation formula, and then the significance of each feature to be trained is calculated by each Pearson correlation value, for example, assuming that each The Pearson correlation values are 0.0001, 0.01, and 0.05, respectively, and the corresponding measurement values for determining each of the significance are 100, 1, and 0.2. The larger the measurement value, the more significant the significance.

Step S30, based on the configuration parameters, select a target training model from the first initial training model and the cyclic training model set;

In this embodiment, it should be noted that the configuration parameters include a model selection strategy.

Based on the configuration parameters, a target training model is selected from the first initial training model and the cyclic training model set. Specifically, based on the model selection strategy, the first initial training model and the cyclic training model From each element of the training model set, a model that best meets the model selection strategy is selected as the target training model.

Wherein, the step of selecting a target training model from the first initial training model and cyclic training model set based on the configuration parameters includes:

Step S31: Obtain a model selection strategy in the parameter configuration, where the model selection strategy includes an AUC value and an AIC value;

In this embodiment, it should be noted that in this embodiment, it should be noted that the AUC value is the criterion for evaluating the training model, and the larger the AUC value is, the better the training model is. Wherein, the AUC value is the area enclosed by the coordinate axis under the ROC (receiver operating characteristic curve) curve, and the value of this area will not be greater than 1, where the ROC curve is based on a A series of different binary classification methods (cutoff value or decision threshold), the true positive rate (sensitivity) is the ordinate, the false positive rate (1-specificity) is the curve drawn on the abscissa, the AIC value is calculated based on the AIC criterion Among them, the AIC criterion is a standard for measuring the goodness of the statistical model.

Step A32, if the model selection strategy is the AUC value, compare the AUC values of the elements in the cyclic training model set, and select the element corresponding to the largest AUC value as the target training model .

In this embodiment, if the model selection strategy is the AUC value, the AUC values of the elements in the cyclic training model set are compared, and the element corresponding to the largest AUC value is selected as the The target training model, specifically, if the model selection strategy is the AUC value, compare the AUC values to obtain the maximum AUC value, and use the training model corresponding to the maximum AUC value as the target training A model, wherein the training model includes a first initial training model and each element in the cyclic training model set.

Step S33, if the model selection strategy is the AIC value, compare the AIC values of the elements in the cyclic training model set, and select the element corresponding to the smallest AIC value as the target training model .

In this embodiment, if the model selection strategy is the AIC value, the AIC value of each element in the cyclic training model set is compared, and the element corresponding to the smallest AIC value is selected as the The target training model, specifically, if the model selection strategy is the AIC value, the AIC values are compared to obtain the minimum AIC value, and the training model corresponding to the minimum AIC value is used as the target training A model, wherein the training model includes a first initial training model and each element in the cyclic training model set.

Step S40: Generate visualization data corresponding to the target training model, and feed back the visualization data to the client.

In this embodiment, it should be noted that the visualization data includes candidate feature visualization data, model selection summary visualization data, and training process visualization data, where the candidate feature is a feature in the feature set to be trained, The model selection summary data includes summary data for model selection of the first initial training model and the model elements in the cyclic training model set.

Generate visualization data corresponding to the target training model, and feed back the visualization data to the client, specifically, generate visualization data corresponding to the acquisition process corresponding to the target training model, wherein the acquisition process includes features Selection process, model training process, model selection process, etc., and then feedback the visualization data to the visualization interface of the client for display to the customer, wherein the feature selection process is the process of selecting features in the feature set to be trained The model training process is a process of training a target model, wherein the target model includes a preset model to be trained, a first initial training model, model elements, etc., and the model selection process is based on a preset model selection strategy The process of selecting the target training model.

Wherein, the client includes a visual interface,

The step of generating visualization data corresponding to the target training model and feeding back the visualization data to the client includes:

Step S41: Obtain candidate feature data, selection summary data, and training process data corresponding to the model selection process of the target training model;

In this embodiment, the model selection process of the target training model includes a model iterative training process, a feature selection process, a model selection process, etc., wherein the feature selection process is a process of removing the feature to be removed, and the model selection process The process of selecting a target training model based on a preset model selection strategy.

Obtain candidate feature data, selection summary data, and training process data corresponding to the model selection process of the target training model, specifically, acquire candidate feature data of the feature selection process and selection summary data of the model selection process in real time And training process data of the model iterative training process.

Step S42: Generate visualization data corresponding to the candidate feature data, the selection summary data, and the training process data, and feed back the visualization data to the visualization interface in real time.

In this embodiment, it should be noted that the visualization data includes graphic data, table data, and the like.

Generate visualization data corresponding to the candidate feature data, the selection summary data, and the training process data, and feed back the visualization data to the visualization interface in real time, specifically, generate the candidate feature data in real time , The selection of the visualization data corresponding to the summary data and the training process data, and the real-time feedback of the visualization data to the visualization interface in real time, wherein the time interval for real-time feedback of the visualization data to the visualization interface The user of the server can be selected by the backward model to set it, and the user of the client can query the visualization data in real time on the client.

In this embodiment, by receiving the configuration parameters sent by the client associated with the server and acquiring the features to be trained, the preset model to be trained is trained based on each of the features to be trained and the configuration parameters to obtain the first initial Training the model, and then calculate the first saliency corresponding to each of the features to be trained, and based on each of the first salience, remove the features to be removed from the features to be trained that meet the preset removal saliency requirements, and then Based on the eliminated features to be trained, the first initial training model is cyclically trained to obtain a cyclic training model set, and then based on the configuration parameters, from the first initial training model and the cyclic training model set Selecting a target training model in, then generating visualization data corresponding to the target training model, and feeding back the visualization data to the client. That is, this embodiment first sends the configuration parameters sent by the client associated with the server and acquires the features to be trained, and based on each of the features to be trained and the configuration parameters, performs a comparison of the preset to be trained The training of the model, the first initial training model is obtained, and the first saliency corresponding to each of the features to be trained is calculated, and then based on each of the first salience, the features to be trained are eliminated in accordance with the preset Remove the features to be removed that require saliency, and then perform cyclic training on the first initial training model based on the removed features to be trained to obtain a cyclic training model set, and then based on the configuration parameters, from the The target training model is selected from the first initial training model and the cyclic training model set, and then the visualization data corresponding to the target training model is generated, and the visualization data is fed back to the client. That is, this embodiment provides a model selection method for the backward selection mode of codeless distributed modeling and visual modeling. The user only needs to set and send the necessary configuration parameters to the server through the client. That is to say, the visual data corresponding to the backward model selection process and the backward model selection result can be fed back, that is, the client and the server are connected to communicate with each other for model modeling, which realizes distributed modeling, which is compared with The modeling of the backward selection mode performed by a stand-alone machine improves the modeling efficiency of the backward selection mode. By generating the visualization data corresponding to the target training model and feeding it back to the client, the visualization modeling is realized, which reduces The ability threshold of modelers is required and the modeling efficiency of the backward selection mode is further improved. In this embodiment, the user only needs to input the necessary model parameters in the visual interface of the client to obtain the corresponding backward model selection results. There is no requirement for the user's code development ability, and thus no code modeling is realized, which further reduces the ability threshold requirement for modelers. Therefore, it solves the high modeling threshold and low efficiency of the backward selection mode in the prior art. Technical issues.

Further, referring to FIG. 3, based on the first embodiment of the present application, in another embodiment of the backward model selection method, in step S20, the cyclic training model set includes one or more model elements, each of which The model element includes the second initial training model,

According to each of the first saliency, the feature to be removed that meets the preset removal saliency requirement is removed from the features to be trained, so as to compare the first initial saliency based on the removed features to be trained. The training model performs cyclic training, and the steps to obtain the cyclic training model set include:

Step C10, based on each of the first saliency and the preset removal saliency requirements, select the feature to be removed among the features to be trained, and remove the feature to be removed;

In this embodiment, it should be noted that the first significance can be determined based on the Pearson correlation value, and when the Pearson correlation value is less than or equal to the preset Pearson correlation threshold, the determination is The feature corresponding to the first saliency does not meet the preset saliency removal requirement, that is, the feature corresponding to the first saliency appears to be significant, when the Pearson correlation value is greater than the preset Pearson correlation threshold When, it is determined that the feature corresponding to the first saliency satisfies the preset saliency removal requirement, that is, the feature corresponding to the first saliency is not significant.

Based on each of the first saliency and the preset removal saliency requirements, select the feature to be removed among the features to be trained, and remove the feature to be removed, specifically, combine each of the first The saliency is compared, the feature with the lowest saliency among the features to be trained is selected as the target feature, and it is judged whether the target feature satisfies the pre-determined saliency requirement, if the target feature meets the pre-determined removal If the saliency requirement is required, the target feature is used as the feature to be eliminated, and the feature to be eliminated is eliminated. If the target feature does not meet the pre-determined saliency requirement for elimination, the current cycle training is ended.

Wherein, the step of selecting the feature to be removed among the features to be trained based on each of the first saliency and the preset removal saliency requirement includes:

Step C11, comparing each of the first saliency, and selecting the feature with the lowest saliency among the features to be trained as the target feature;

In this embodiment, each of the first saliency is compared, and the feature with the lowest saliency is selected as the target feature among the features to be trained. Specifically, the first saliency is selected as a target feature. A comparison to obtain the least significant feature of each of the features to be trained corresponding to each of the saliency, that is, to obtain the feature with the highest Pearson correlation value, that is, in each of the features to be trained The least significant feature is selected as the target feature.

Step C12, comparing the target saliency of the target feature with a preset saliency rejection threshold;

Step C13: If the target significance is less than the preset rejection significance threshold, it is determined that the target feature meets the preset rejection significance requirement, and the target feature is taken as the feature to be rejected.

In this embodiment, the target saliency of the target feature is compared with a preset rejection saliency threshold, and if the target saliency is less than the preset rejection saliency threshold, the target feature is determined Meet the preset saliency requirement for rejection, and use the target feature as the feature to be rejected. Specifically, the target saliency of the target feature is compared with a preset saliency threshold, wherein the target The saliency is the first saliency of the target feature. If the target saliency is lower than the preset saliency threshold, the target feature meets the preset saliency removal requirement, that is, the The target feature is not significant, and then the target feature is used as the feature to be eliminated. If the target significance is higher than or equal to the preset significance threshold, the target feature does not satisfy the preset Excluding the significance requirement, that is, the target feature is significant, then this cycle training is ended.

Step C20, training the first initial training model based on the eliminated features to be trained to obtain the second initial training model.

In this embodiment, it should be noted that the cyclic training model set includes one or more model elements.

Based on the eliminated features to be trained, the first initial training model is trained to obtain the second initial training model. Specifically, the feature data of the eliminated features to be trained is input into the A first initial training model to perform an iterative training update on the first initial training model until the updated first initial training model satisfies a preset training completion judgment condition to obtain the updated first initial training model That is, the second initial training model is obtained, wherein the preset training completion judgment condition includes reaching the maximum number of iterations and reaching the minimum convergence error.

Step C30: Calculate the second saliency of each feature to be trained after culling, and based on each of the second saliency, remove again from each feature to be trained after culling that meets the preset removal saliency Other required features to be removed;

In this embodiment, the second saliency of each feature to be trained after being removed is calculated, and based on each of the second saliency, the removal of each feature to be trained after removal is again consistent with the preset The other features to be removed that require saliency are removed, specifically, the chi-square value wald of each feature to be trained after removal is recalculated, and based on the recalculated chi-square value wald and each removed feature. The degrees of freedom of the features to be trained are calculated, and the second saliency of each feature to be trained after being removed is calculated, and based on each of the second saliency, it is determined whether there is any feature that satisfies the preset after being removed. Remove the feature to be removed that requires saliency. If there are other features to be removed that meet the preset requirement of removal saliency among the removed features to be trained, the other features to be removed will be removed again. If there are no other features to be eliminated that meet the pre-determined saliency requirement for elimination among the features to be trained, the current cycle training is ended.

Step C40: Perform cyclic training on the second initial training model based on each of the features to be trained after being removed again, to obtain one or more of the model elements, until the feature to be trained does not exist in each of the features to be trained. Remove features.

In this embodiment, based on each of the features to be trained after being removed again, the second initial training model is cyclically trained to obtain one or more of the model elements until there is no feature in each of the features to be trained The features to be removed, specifically, based on the features to be trained after being removed again, the second initial training model is iteratively trained and updated until the second initial training model reaches the training completion judgment condition, and the update is obtained The latter second initial training model, that is, one of the model elements is obtained, and the search and elimination of the features to be eliminated are re-circulated, and the bone-setting training update of the cyclically updated second initial training model is performed, Obtain one or more model elements, until there is no feature to be removed that meets the preset removal significance requirement among the features to be trained, then this cyclic training is ended, and then a cyclic training model set is obtained, as shown in Figure 4 This embodiment is a schematic diagram of the flow of backward model selection in combination with the first embodiment, where the features in the model are each of the features to be trained, and the training model is the preset model to be trained or the pre-trained model. It is assumed that the model to be trained, such as the first initial training model or other model elements, etc., and the threshold value is the preset significance threshold for rejection.

In this embodiment, based on each of the first saliency and the preset removal saliency requirement, the feature to be removed from the features to be trained is selected, and the feature to be removed is removed, and then based on each removed feature For the feature to be trained, the first initial training model is trained to obtain the second initial training model, and then the second saliency of each feature to be trained after being eliminated is calculated, and based on each of the second Saliency, among the features to be trained after being removed, other features to be removed that meet the pre-determined saliency requirement are removed again, and then based on the features to be trained after being removed again, the first 2. The initial training model performs cyclic training to obtain one or more of the model elements until the feature to be removed does not exist in each feature to be trained. That is, in this embodiment, by calculating the saliency of each of the features to be trained, the features to be eliminated in each feature to be trained are eliminated one by one, and the first initial training model is analyzed based on the features to be trained after each elimination. The training update is performed until the feature to be removed does not exist in each feature to be trained, the cyclic training model set is obtained, and the model selection of the backward selection mode can be performed based on the cyclic training model set, that is, By calculating and analyzing the corresponding saliency of each feature to be trained, the feature to be removed for each feature to be trained is gradually eliminated to perform cyclic training on the first initial training model to obtain a cyclic training model set, and then to achieve no code The model selection of the backward selection mode of distributed modeling and visual modeling lays the foundation, that is, it lays a foundation for solving the technical problems of high threshold and low efficiency of backward selection mode modeling in the prior art.

Further, referring to FIG. 5, based on the first embodiment of the present application, in another embodiment of the forward model selection method, the forward model selection method is applied to the client, and the forward model selection method includes:

Step A10: Receive a model selection task, and send configuration parameters corresponding to the model selection task to the server associated with the client, so that the server can perform model selection based on the configuration parameters to obtain a target training model , And obtain the visualization data corresponding to the target training model, so as to send the visualization data to the client;

In this embodiment, it should be noted that the model selection task includes target model requirements, the target model requirements are determined by the configuration parameters, and the configuration parameters include large iteration coefficients, minimum convergence errors, model selection modes, etc. parameter.

Receive the model selection task, and send the configuration parameters corresponding to the model selection task to the server associated with the client, so that the server can make model selection based on the configuration parameters, obtain the target training model, and obtain The visualization data corresponding to the target training model is sent to the client, specifically, the model selection task is received, and the configuration parameters corresponding to the model selection task are matched in a preset local database or determined by The user sets the configuration parameters by himself based on the model selection task, and further, sends the configuration parameters to the server associated with the client, so that the server can perform a preset initialization based on the configuration parameters. The training update of the model, the model to be trained is obtained, and the cyclic training update is performed on the model to be trained to obtain one or more models to be selected, that is, the cyclic training model set is obtained, and the model to be selected is selected in each of the models to be selected. The model of the preset model selection strategy is used as the target training model, and the process data corresponding to the target training model is converted into the visualization data and fed back to the client, where the visualization data includes candidate feature visualization data and models Select and summarize visualization data and model training process visualization data, where the candidate features are each of the features to be trained, and the model selection summary data includes performing model elements in the cyclic training model set based on a preset model selection strategy. Summary data for model selection.

Step A20: Receive the visualization data fed back by the server, and display the visualization data on a preset visualization interface.

In this embodiment, it should be noted that the client can query the visualization data corresponding to the process data of the server in real time on the preset visualization interface, and it can be in the process of model selection or model selection. After the selection is completed, the process data is inquired, and the client is in communication with the server.

In this embodiment, a model selection task is received, and the configuration parameters corresponding to the model selection task are sent to the server associated with the client, so that the server can perform model selection based on the configuration parameters to obtain target training Model, and obtain the visualization data corresponding to the target training model to send the visualization data to the client, and then receive the visualization data fed back by the server, and set the visualization data in a preset visualization The interface is displayed. That is, this implementation provides a model selection method for codeless distributed modeling and visual modeling. The user only needs to set and send the necessary configuration parameters to the server through the client, and the server can feed back the corresponding visual data That is, this embodiment implements distributed modeling, improves the modeling efficiency during model selection, and the model selection process does not have any code development capability requirements for users, which reduces the ability threshold requirements for modelers. And because the server can convert the process data corresponding to the target training model into visualized data and feed it back to the client, it further reduces the ability threshold requirements for modelers, and the visualized data is convenient for modelers to understand and read. In turn, the modeling efficiency of modelers can be further improved, and therefore, the technical problems of high threshold and low efficiency of forward selection model modeling in the prior art are solved.

Referring to FIG. 6, FIG. 6 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present application.

As shown in FIG. 6, the backward model selection device may include a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to implement connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as a magnetic disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

In an embodiment, the backward model selection device may further include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, and so on. The rectangular user interface may include a display screen (Display) and an input sub-module such as a keyboard (Keyboard), and the optional rectangular user interface may also include a standard wired interface and a wireless interface. The network interface can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

Those skilled in the art can understand that the structure of the backward model selection device shown in FIG. 6 does not constitute a limitation on the backward model selection device, and may include more or less components than shown in the figure, or a combination of certain components, Or different component arrangements.

As shown in FIG. 6, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, and a backward model selection program. The operating system is a program that manages and controls the hardware and software resources of the backward model selection device, and supports the operation of the backward model selection program and other software and/or programs. The network communication module is used to realize the communication between the components in the memory 1005 and the communication with other hardware and software in the backward model selection system.

In the backward model selection device shown in FIG. 6, the processor 1001 is configured to execute the backward model selection program stored in the memory 1005 to implement the steps of the backward model selection method described in any one of the foregoing items.

The specific implementation of the backward model selection device of the present application is basically the same as the foregoing embodiments of the backward model selection method, and will not be repeated here.

An embodiment of the present application also provides a backward model selection device. The backward model selection device is applied to a server, and the backward model selection device includes:

The first training module is configured to receive the configuration parameters sent by the client associated with the server and obtain the features to be trained, and train a preset model to be trained based on each of the features to be trained and the configuration parameters , To obtain the first initial training model;

The second training module is used to calculate the first saliency corresponding to each of the features to be trained, and based on each of the first saliences, to remove the features that meet the preset removal saliency requirements from the features to be trained The features to be eliminated, to perform cyclic training on the first initial training model based on each of the features to be trained after culling, to obtain a cyclic training model set;

A selection module for selecting a target training model from the first initial training model and a set of cyclic training models based on the configuration parameters;

The feedback module is used for generating the visualization data corresponding to the target training model, and feeding back the visualization data to the client.

In an embodiment, the second training module includes:

The first culling sub-module is configured to select the feature to be removed among the features to be trained based on each of the first saliency and the preset saliency removal requirement, and to remove the feature to be removed ；

A training sub-module, configured to train the first initial training model based on the eliminated features to be trained to obtain the second initial training model;

The second culling sub-module is used to calculate the second saliency of each feature to be trained after being removed, and based on each of the second saliency, remove the coincidence again from each feature to be trained after being removed Other features to be removed that are required to be removed by the preset saliency;

The cyclic training sub-module is used to perform cyclic training on the second initial training model based on each of the features to be trained after being removed again, to obtain one or more of the model elements, until each feature to be trained The feature to be removed does not exist in.

In an embodiment, the selection submodule includes:

The first comparison unit is configured to compare each of the first saliency, and select the feature with the lowest saliency among the features to be trained as the target feature;

The second comparison unit is used to compare the target significance of the target feature with a preset significance threshold for rejection;

The determining unit is configured to determine that if the target significance is less than the preset rejection significance threshold, determine that the target feature meets the preset rejection significance requirement, and use the target feature as the pending Remove features.

In an embodiment, the second training module further includes:

The first calculation sub-module is used to calculate the chi-square value wald of each of the features to be trained;

The second calculation sub-module is used for calculating each of the first saliency based on each of the chi-square value wald and the degrees of freedom of each of the features to be trained.

In an embodiment, the first training module includes:

A training update sub-module for inputting the feature data corresponding to each of the features to be trained into the preset model to be trained, so as to train and update the preset model to be trained;

The first judging sub-module is used to judge whether the updated preset model to be trained satisfies the training completion judging condition, and if the updated preset to be trained model satisfies the training completion judging condition, then Obtaining the first initial training model;

The second judgment sub-module is configured to continue to perform iterative training updates on the preset to-be-trained model if the updated preset to-be-trained model does not satisfy the training completion judgment condition until the updated all-in-one model The preset model to be trained satisfies the training completion judgment condition.

In an embodiment, the selection module includes:

The first obtaining sub-module is configured to obtain the model selection strategy in the parameter configuration, wherein the model selection strategy includes an AUC value and an AIC value;

The first comparison sub-module is configured to compare the AUC value of each element in the cyclic training model set if the model selection strategy is the AUC value to select the largest corresponding AUC value As the target training model;

The second comparison sub-module is used to compare the AIC value of each element in the cyclic training model set if the model selection strategy is the AIC value to select the smallest corresponding AIC value As the target training model.

In an embodiment, the feedback module includes:

The second acquisition sub-module is used to acquire the candidate feature data, selection summary data, and training process data corresponding to the backward model selection process of the target training model;

A generating sub-module is used to generate the visualization data corresponding to the candidate feature data, the selection summary data, and the training process data, and feed back the visualization data to the visualization interface in real time.

The specific implementation of the backward model selection device of the present application is basically the same as the foregoing embodiments of the backward model selection method, and will not be repeated here.

To achieve the foregoing objective, an embodiment of the present application also provides a backward model selection device, the backward model selection device is applied to a client, and the backward model selection device includes:

The sending module is configured to receive the model selection task, and send the configuration parameters corresponding to the model selection task to the server associated with the client, so that the server can use the configuration parameters and the acquired configuration parameters. Performing model selection on training features, obtaining a target training model, and obtaining visualization data corresponding to the target training model, so as to send the visualization data to the client;

The receiving module is configured to receive the visualization data fed back by the server, and display the visualization data on a preset visualization interface.

The specific implementation of the backward model selection device of the present application is basically the same as the foregoing embodiments of the backward model selection method, and will not be repeated here.

The embodiments of the present application provide a readable storage medium, and the readable storage medium stores one or more programs, and the one or more programs may also be executed by one or more processors for implementation The steps of the backward model selection method described in any one of the above.

The specific implementation of the readable storage medium of the present application is basically the same as each embodiment of the backward model selection method described above, and will not be repeated here.

The above are only the preferred embodiments of the application, and do not limit the scope of the patent for this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of the application, or directly or indirectly applied to other related technical fields , The same reason is included in the scope of patent processing of this application.

Claims (20)

1. A method for selecting a backward model, wherein the method for selecting a backward model is applied to a server, and the method for selecting a backward model includes:

   Receiving configuration parameters sent by the client associated with the server and acquiring features to be trained, and training a preset model to be trained based on each of the features to be trained and the configuration parameters to obtain a first initial training model;

   Calculate the first saliency corresponding to each of the features to be trained, and based on each of the first saliency, eliminate the features to be removed that meet the preset saliency requirements for removal from the features to be trained, so as to be based on the removed features. Each of the features to be trained performs cyclic training on the first initial training model to obtain a cyclic training model set;

   Based on the configuration parameters, selecting a target training model from the first initial training model and the cyclic training model set;

   Generate visualization data corresponding to the target training model, and feed back the visualization data to the client.
2. The backward model selection method according to claim 1, wherein the cyclic training model set includes one or more model elements, and each of the model elements includes a second initial training model,

   According to each of the first saliency, the feature to be removed that meets the preset removal saliency requirement is removed from the features to be trained, so as to compare the first initial saliency based on the removed features to be trained. The training model performs cyclic training, and the steps to obtain the cyclic training model set include:

   Based on each of the first saliency and the preset removal saliency requirement, select the feature to be removed from the features to be trained, and remove the feature to be removed;

   Training the first initial training model based on each of the features to be trained after being eliminated to obtain the second initial training model;

   Calculate the second saliency of each feature to be trained after culling, and based on each of the second saliency, remove other features that meet the preset removal saliency requirements from the features to be trained after removal. The features to be removed;

   Based on each of the features to be trained after being removed again, the second initial training model is cyclically trained to obtain one or more of the model elements until the feature to be removed does not exist in each of the features to be trained.
3. 3. The backward model selection method according to claim 2, wherein the selected feature of the feature to be removed from the feature to be trained is based on each of the first saliency and the preset removal saliency requirement The steps include:

   Comparing each of the first saliency, and selecting the feature with the lowest saliency among the features to be trained as the target feature;

   Comparing the target saliency of the target feature with a preset saliency rejection threshold;

   If the target significance is less than the preset rejection significance threshold, it is determined that the target feature meets the preset rejection significance requirement, and the target feature is taken as the feature to be rejected.
4. The backward model selection method according to claim 1, wherein the step of calculating the first saliency corresponding to each of the features to be trained comprises:

   Calculating the chi-square value wald of each of the features to be trained;

   Based on each of the chi-square value wald and the degrees of freedom of each of the features to be trained, each of the first saliences is calculated.
5. 5. The backward model selection method according to claim 1, wherein the configuration parameters include training completion judgment conditions, and the features to be trained include one or more pieces of feature data;

   The step of training a preset model to be trained based on each of the features to be trained and the configuration parameters, and obtaining a first initial training model includes:

   Input the feature data corresponding to each of the features to be trained into the preset model to be trained, so as to train and update the preset model to be trained;

   Judging whether the updated preset to-be-trained model satisfies the training completion judgment condition, and if the updated preset to-be-trained model satisfies the training completion judgment condition, the first initial training model is obtained;

   If the updated preset to-be-trained model does not meet the training completion judgment condition, continue to perform iterative training updates on the preset to-be-trained model until the updated preset to-train model satisfies the training Complete the judgment condition.
6. 5. The backward model selection method according to claim 1, wherein the step of selecting a target training model from the first initial training model and a set of cyclic training models based on the configuration parameters comprises:

   Acquiring a model selection strategy in the parameter configuration, where the model selection strategy includes an AUC value and an AIC value;

   If the model selection strategy is the AUC value, compare the AUC values of the elements in the cyclic training model set, and select the element corresponding to the largest AUC value as the target training model;

   If the model selection strategy is the AIC value, the AIC values of the elements in the cyclic training model set are compared, and the element corresponding to the smallest AIC value is selected as the target training model.
7. The backward model selection method according to claim 1, wherein the client includes a visual interface,

   The step of generating visualization data corresponding to the target training model and feeding back the visualization data to the client includes:

   Acquiring candidate feature data, selection summary data, and training process data corresponding to the model selection process of the target training model;

   Generate visualization data corresponding to the candidate feature data, the selection summary data, and the training process data, and feed back the visualization data to the visualization interface in real time.
8. 5. The backward model selection method according to claim 5, wherein the training completion judgment condition includes reaching a minimum convergence error and reaching a maximum number of iterations.
9. The backward model selection method according to claim 4, wherein the insignificant feature refers to a feature whose significance is lower than a preset significance threshold among the features to be trained, wherein the significance is based on The chi-square value wald and the degree of freedom of the feature to be trained are acquired, wherein the degree of freedom is related to the value of the feature to be trained.
10. The backward model selection method of claim 1, wherein the first significance is determined based on a Pearson correlation value, when the Pearson correlation value is less than or equal to a preset Pearson correlation valve Value, it is determined that the feature corresponding to the first saliency does not meet the preset saliency removal requirement.
11. 5. The backward model selection method according to claim 4, wherein said calculating the chi-square value wald of each of the features to be trained comprises:

    The feature data representation matrix corresponding to each feature to be trained is substituted into the preset chi-square value wald calculation formula, and the chi-square value wald corresponding to each feature to be trained is calculated in parallel, wherein the preset chi-square value wald The calculation formula is as follows:

    

    among them,

    

    S is the chi-square value wald, X is the feature data corresponding to the feature to be trained, where X includes n pieces of data, and each piece of data includes k values, and the feature set to be trained X is divided into a first model feature set X ₀ and the second model feature set X1, X ₀ includes n pieces of data, each piece of data includes (kt) numeric values, X ₁ includes n pieces of data, each piece of data includes t numeric values, and X trains the preset to be trained The model parameter obtained by the model is θ, the data set corresponding to the target output of the model to be trained is Y, Y includes n pieces of data, and Y corresponds to the existence prediction probability P, P includes n probabilities (p ₁ , p ₂ ,... , P _n-1 , p _n ), C is a matrix of t*k, and h is a vector of k*1.
12. The backward model selection method according to claim 4, wherein the second saliency of each of the features to be trained after the culling of the calculation comprises:

    Recalculate the chi-square value wald of each feature to be trained after removal, and calculate each of the removed chi-square values wald based on the recalculated chi-square value wald and the degrees of freedom of each feature to be trained after removal The second significance of the feature to be trained.
13. 7. The backward model selection method according to claim 6, wherein the AUC value is the area under the ROC curve enclosed by the coordinate axis, and the value of the area is less than or equal to 1.
14. A backward model selection method, wherein the backward model selection method is applied to a client, and the backward model selection method includes:

    Receive a model selection task, and send the configuration parameters corresponding to the model selection task to the server associated with the client, so that the server performs model selection based on the configuration parameters and the acquired features to be trained to obtain A target training model, and obtaining visualization data corresponding to the target training model, so as to send the visualization data to the client;

    The visualization data fed back by the server is received, and the visualization data is displayed on a preset visualization interface.
15. A backward model selection device, wherein the backward model selection device includes a memory, a processor, and a program stored on the memory for implementing the backward model selection method,

    The memory is used to store a program for implementing the backward model selection method;

    The processor is configured to execute a program that implements the backward model selection method to implement the backward model selection method. The backward model selection method includes: receiving configuration parameters sent by a client associated with the server and obtaining Training features to be trained, and training a preset model to be trained based on each of the features to be trained and the configuration parameters to obtain a first initial training model;

    Calculate the first saliency corresponding to each of the features to be trained, and based on each of the first saliency, eliminate the features to be removed that meet the preset saliency requirements for removal from the features to be trained, so as to be based on the removed features. Each of the features to be trained performs cyclic training on the first initial training model to obtain a cyclic training model set;

    Based on the configuration parameters, selecting a target training model from the first initial training model and the cyclic training model set;

    Generate visualization data corresponding to the target training model, and feed back the visualization data to the client.
16. The backward model selection device according to claim 15, wherein the processor is configured to execute a program for implementing the backward model selection method to implement the following steps: the cyclic training model set includes one or more model elements , Each of the model elements includes a second initial training model,

    According to each of the first saliency, the feature to be removed that meets the preset removal saliency requirement is removed from the features to be trained, so as to compare the first initial saliency based on the removed features to be trained. The training model performs cyclic training, and the steps to obtain the cyclic training model set include:

    Based on each of the first saliency and the preset removal saliency requirement, select the feature to be removed from the features to be trained, and remove the feature to be removed;

    Training the first initial training model based on each of the features to be trained after being eliminated to obtain the second initial training model;

    Calculate the second saliency of each feature to be trained after culling, and based on each of the second saliency, remove other features that meet the preset removal saliency requirements from the features to be trained after removal. The features to be removed;

    Based on each of the features to be trained after being removed again, the second initial training model is cyclically trained to obtain one or more of the model elements until the feature to be removed does not exist in each of the features to be trained.
17. The backward model selection device according to claim 16, wherein the processor is configured to execute a program that implements the backward model selection method to implement the following steps:

    The step of selecting the feature to be removed among the features to be trained based on each of the first saliency and the preset removal saliency requirement includes:

    Comparing each of the first saliency, and selecting the feature with the lowest saliency among the features to be trained as the target feature;

    Comparing the target saliency of the target feature with a preset saliency rejection threshold;

    If the target significance is less than the preset rejection significance threshold, it is determined that the target feature meets the preset rejection significance requirement, and the target feature is taken as the feature to be rejected.
18. 15. The backward model selection device according to claim 15, wherein the processor is configured to execute a program that implements the backward model selection method to implement the following steps:

    The step of calculating the first saliency corresponding to each of the features to be trained includes:

    Calculating the chi-square value wald of each of the features to be trained;

    Based on each of the chi-square value wald and the degrees of freedom of each of the features to be trained, each of the first saliences is calculated.
19. 15. The backward model selection device according to claim 15, wherein the processor is configured to execute a program that implements the backward model selection method to implement the following steps:

    The configuration parameters include training completion judgment conditions, and the features to be trained include one or more pieces of feature data;

    The step of training a preset model to be trained based on each of the features to be trained and the configuration parameters, and obtaining a first initial training model includes:

    Input the feature data corresponding to each of the features to be trained into the preset model to be trained, so as to train and update the preset model to be trained;

    Judging whether the updated preset to-be-trained model satisfies the training completion judgment condition, and if the updated preset to-be-trained model satisfies the training completion judgment condition, the first initial training model is obtained;

    If the updated preset to-be-trained model does not meet the training completion judgment condition, continue to perform iterative training updates on the preset to-be-trained model until the updated preset to-train model satisfies the training Complete the judgment condition.
20. A readable storage medium, wherein a program for implementing the backward model selection method is stored on the readable storage medium, and the program for implementing the backward model selection method is executed by a processor to implement claims 1 to 13 or Steps of the backward model selection method described in any one of 14.

Images