extensible.extensions.default

Classes

EvalState([visualize])	Exposes the following fixtures:
Setup()	Basic train manager setup.

class extensible.extensions.default.Setup

Bases: Extension

Basic train manager setup.

pre_run(train_manager, standalone_eval)

Basic train manager setup. This hook

1. moves the model to the device,

2. builds the optimizer if it was provided as a callable, passing in all model parameters and

post_load_checkpoint(checkpoint_loaded, fixtures, train_manager)

Initializes weights if no checkpoint was loaded.

class extensible.extensions.default.EvalState(visualize: bool = True)

Bases: Extension

Exposes the following fixtures:

Fixtures

num_batches

The cumulative number of batches currently visited during an evaluation epoch.

num_samples

The cumulative number of samples currently visited during an evaluation epoch.

cum_loss

The sum of all losses for all batches currently visited during an evalution epoch.

Adds the following visualization at the end of the epoch:

• Cumulative loss divided by the total number of batches in the epoch.