[Feature] To track the training callback details

[mahimairaja]

Feature request

What is the expected behavior? To track the training callback details

What is motivation or use case for adding/changing the behavior? In tensorflow and pytorch we could easily track the performance of the model in a variable history which is a sort of dict like.

How should this be implemented in your opinion? An class similar to History from tf.keras.callback can be written with similar logic

Are you willing to work on this yourself? yes

[Optimox]

I am not sure I understand, call backs are already implemented. Moreover, you can access model.history to get information about the training

[mahimairaja]

I just trained the model using the below line:

history = best_model.fit(xtrain, ytrain, patience=10, \
                max_epochs=50, eval_set=[(xtrain, ytrain)])

And I got the below values in a dict when I access with best_model.history.history.keys()

['loss', 'lr', 'val_0_accuracy']

But tracking values like - acc, val_acc, loss, val_loss could be more feasible as compared to tensorflow

[Optimox]

The history will contain all metrics tracked on the different evaluations sets given:

• add (x_valid, y_valid) to your eval_set
• add more metric as the loss you are willing to monitor in your eval_metrics

[Optimox]

Please feel free to reopen if needed.