Using callbacks to get the most out of your training cycle

So far in this book, we have kicked off the training process for a fastai model by applying fit_one_cycle or fine_tune to the learner object and have then just let the training run through the specified number of epochs. For many of the models we have trained in this book, this approach has been good enough, particularly for models where each epoch takes a long time and we only train for one or two epochs. But what about models where we want to train the model for 10 or more epochs? If we simply let the training process go to the end, we face the problem shown in the training results shown in Figure 8.30. In Figure 8.30, we see the result of training a tabular model for 10 epochs with metric set to accuracy:

Figure 8.30 – Results from a 10-epoch run training a model with tabular data

The goal of this training process is to get a model with the best accuracy. With this goal in mind, there are...