A prediction model training method and device, storage medium, and electronic equipment

Abstract

The embodiment of the invention discloses a prediction model training method and device, a storage medium, and electronic equipment. Wherein the method step includes obtaining an intermediate gradient vector of each first sample data in a plurality of first sample data based on a prediction model using a first model parameter, where the first sample data includes N features, and N is a positive integer; Obtain the weight corresponding to each of the N features according to the effective feature values ​​of multiple second sample data, wherein the second sample data and the first sample data belong to the same sample data set; according to the weight corresponding to each feature , the intermediate gradient vector, the total number of samples of the multiple first sample data, and obtain the target gradient vectors of the multiple first sample data; update the first model parameters according to the target gradient vectors to obtain the second model parameters. By adopting the present application, the influence of the feature in updating the model parameters can be increased, thereby improving the accuracy of the prediction model.

Description

technical field [0001] The present invention relates to the technical field of predictive models, in particular to a predictive model training method and device, a storage medium, and electronic equipment. Background technique [0002] Machine learning-based predictive model training usually uses a sample data set to train a predictive model suitable for the task; wherein, the sample data set usually contains eigenvalues ​​of multi-dimensional features. The existing multi-sample-based forecasting model training is to calculate the respective gradient values ​​for each sample data in multiple sample data, and then average the sum of all gradient values ​​according to the total number of multiple sample data, and calculate the gradient value according to the average gradient value to update the predictive model parameters. However, not all eigenvalues ​​of the sample data set are effective, which will make the number of effective eigenvalues ​​in each dimension feature differ...

AI Technical Summary

Problems solved by technology

However, not all eigenvalues ​​of the sample data set are effective, which will make the number of effective eigenvalues ​​in each dimension feature different. For example, the effective eigenvalues ​​of individual features are relatively sparse, so the After averaging the sum of the gradient values ​​of , the average gradient value corresponding to the feature of this dimension is extremely small, so that it will not affect the update of the model parameters of the prediction model, which reduces the influence of this feature in the update of model parameters, and also reduces The accuracy of the predictive model

Images