Early Stopping for PyTorch

Early stopping is a form of regularization used to avoid overfitting on the training dataset. Early stopping keeps track of the validation loss, if the loss stops decreasing for several epochs in a row the training stops. The EarlyStopping class in pytorchtool.py is used to create an object to keep track of the validation loss while training a PyTorch model. It will save a checkpoint of the model each time the validation loss decrease. We set the patience argument in the EarlyStopping class to how many epochs we want to wait after the last time the validation loss improved before breaking the training loop. There is a simple example of how to use the EarlyStopping class in the MNIST_Early_Stopping_example notebook.

Underneath is a plot from the example notebook, which shows the last checkpoint made by the EarlyStopping object, right before the model started to overfit. It had patience set to 20.

Usage

You can run this project directly in the browser by clicking this button: , or you can clone the project to your computer and install the required pip packages specified in the requirements text file.

pip install -r requirements.txt

References

The EarlyStopping class in pytorchtool.py is inspired by the ignite EarlyStopping class.