Callbacks

Eztorch supports callbacks for trainer. It can takes callbacks from pytorch lightning or custom ones.

Custom callbacks

class eztorch.callbacks.OnlineEvaluator(optimizer, classifier, input_name='h', precision=32)[source]

Attaches a classifier to evaluate a specific representation from the model during training.

Parameters:
  • optimizer (DictConfig) – Config to instantiate an optimizer and optionally a scheduler.

  • classifier (DictConfig) – Config to instantiate a classifier.

  • input_name (str, optional) – Name of the representation to evaluate from the model outputs.

    Default: 'h'

  • precision (int, optional) – Precision for the classifier that must match the model, if \(16\) use automatic mixed precision.

    Default: 32

Example:

optimizer = {...} # config to build an optimizer
classifier = {...} # config to build a classifier
trainer = Trainer(callbacks=[OnlineEvaluator(optimizer, classifier)])