avalanche.training.plugins.LRSchedulerPlugin

class avalanche.training.plugins.LRSchedulerPlugin(scheduler, reset_scheduler=True, reset_lr=True, metric=None, step_granularity: Literal['epoch', 'iteration'] = 'epoch', first_epoch_only=False, first_exp_only=False)[source]

Learning Rate Scheduler Plugin.

This plugin manages learning rate scheduling inside of a strategy using the PyTorch scheduler passed to the constructor. The step() method of the scheduler is called after each training epoch or iteration.

Metric-based schedulers (like ReduceLROnPlateau) are supported as well.

__init__(scheduler, reset_scheduler=True, reset_lr=True, metric=None, step_granularity: Literal['epoch', 'iteration'] = 'epoch', first_epoch_only=False, first_exp_only=False)[source]

Creates a LRSchedulerPlugin instance.

Parameters
  • scheduler – a learning rate scheduler that can be updated through a step() method and can be reset by setting last_epoch=0.

  • reset_scheduler – If True, the scheduler is reset at the end of the experience. Defaults to True.

  • reset_lr – If True, the optimizer learning rate is reset to its original value. Default to True.

  • metric – the metric to use. Must be set when using metric-based scheduling (like ReduceLROnPlateau). Only “train_loss” and “val_loss” are supported at the moment. Beware that, when using “val_loss”, the periodic evaluation flow must be enabled in the strategy. By default, the eval_every parameter of the base strategy is -1, which means that the validation set is never evaluated. Set that value to 1 to obtain the correct results. Also, when using metric=”val_loss”, remember to pass a proper validation stream to the strategy train method, otherwise the periodic evaluation stream will use the training set to compute the validation loss.

  • step_granularity – defines how often the scheduler’s step() method will be called. Defaults to ‘epoch’. Valid values are ‘epoch’ and ‘iteration’.

  • first_epoch_only – if True, the scheduler will only be stepped in the first epoch of each training experience. This is not mutually exclusive with first_exp_only: by setting both values to True, the scheduler will be stepped only in the very first epoch of the whole training stream.

  • first_exp_only – if True, the scheduler will only be considered in the first training experience.

Methods

__init__(scheduler[, reset_scheduler, ...])

Creates a LRSchedulerPlugin instance.

after_backward(strategy, *args, **kwargs)

Called after criterion.backward() by the BaseTemplate.

after_eval(strategy, **kwargs)

Called after eval by the BaseTemplate.

after_eval_dataset_adaptation(strategy, ...)

Called after eval_dataset_adaptation by the BaseTemplate.

after_eval_exp(strategy, *args, **kwargs)

Called after eval_exp by the BaseTemplate.

after_eval_forward(strategy, *args, **kwargs)

Called after model.forward() by the BaseTemplate.

after_eval_iteration(strategy, **kwargs)

Called after the end of an iteration by the BaseTemplate.

after_forward(strategy, *args, **kwargs)

Called after model.forward() by the BaseTemplate.

after_train_dataset_adaptation(strategy, ...)

Called after train_dataset_adapatation by the BaseTemplate.

after_training(strategy, **kwargs)

Called after train by the BaseTemplate.

after_training_epoch(strategy, **kwargs)

Called after train_epoch by the BaseTemplate.

after_training_exp(strategy, **kwargs)

Called after train_exp by the BaseTemplate.

after_training_iteration(strategy, **kwargs)

Called after the end of a training iteration by the BaseTemplate.

after_update(strategy, *args, **kwargs)

Called after optimizer.update() by the BaseTemplate.

before_backward(strategy, *args, **kwargs)

Called before criterion.backward() by the BaseTemplate.

before_eval(strategy, *args, **kwargs)

Called before eval by the BaseTemplate.

before_eval_dataset_adaptation(strategy, ...)

Called before eval_dataset_adaptation by the BaseTemplate.

before_eval_exp(strategy, *args, **kwargs)

Called before eval_exp by the BaseTemplate.

before_eval_forward(strategy, *args, **kwargs)

Called before model.forward() by the BaseTemplate.

before_eval_iteration(strategy, *args, **kwargs)

Called before the start of a training iteration by the BaseTemplate.

before_forward(strategy, *args, **kwargs)

Called before model.forward() by the BaseTemplate.

before_train_dataset_adaptation(strategy, ...)

Called before train_dataset_adapatation by the BaseTemplate.

before_training(strategy, **kwargs)

Called before train by the BaseTemplate.

before_training_epoch(strategy, *args, **kwargs)

Called before train_epoch by the BaseTemplate.

before_training_exp(strategy, *args, **kwargs)

Called before train_exp by the BaseTemplate.

before_training_iteration(strategy, **kwargs)

Called before the start of a training iteration by the BaseTemplate.

before_update(strategy, *args, **kwargs)

Called before optimizer.update() by the BaseTemplate.