Model training method and device and medium

Abstract

The invention discloses a model training method. The method comprises the steps: obtaining multiple groups of hyper-parameters, and constructing models by utilizing each group of hyper-parameters; respectively training the plurality of constructed models by using a training set, and verifying the model being trained by using a verification set; in response to triggering early stop, obtaining evaluation parameters generated when the model being trained is verified, and obtaining a standard parameter from the plurality of evaluation parameters; judging whether the standard parameter is greater than a threshold; in response to the condition that the standard parameter is not greater than the threshold, obtaining a reciprocal of a loss function value corresponding to the currently trained model, and determining a plurality of models which continue to be trained from the models being trained according to the obtained reciprocal of the loss function value and the corresponding evaluation parameters; and in response to the situation that the number of the continuously trained models is greater than 1, returning to the step of training until the number of the continuously trained models isequal to 1. The invention further discloses computer equipment and a readable storage medium.

Description

technical field [0001] The invention relates to the field of hyperparameter search, in particular to a model training method, system, device and storage medium. Background technique [0002] When using the hyperparameter tuning scheme on the model, the corresponding evaluation parameters are often used as the standard for evaluating the network training effect and judging the quality of the hyperparameter combination. If the early stopping algorithm is used in hyperparameter tuning, the steps are roughly as follows: use a certain sampling method to determine one or a group of hyperparameter combinations, and obtain the target detection model corresponding to each group of hyperparameter combinations; according to the specific parameters of the early stopping algorithm Requirements, the model is trained for several rounds, and the corresponding evaluation value of the verification set is obtained. In some algorithms, the quality of the model may be judged based on the existi...

AI Technical Summary

Problems solved by technology

In either case, the early stopping strategy will be very sloppy, and it is very likely that some training with good final training effect will be stopped in advance, which will definitely affect the final training accuracy and hyperparameter selection

Images