flair.trainers.plugins.LinearSchedulerPlugin#

class flair.trainers.plugins.LinearSchedulerPlugin(warmup_fraction)View on GitHub #

Plugin for LinearSchedulerWithWarmup.

Methods

`__init__`(warmup_fraction)	Initialize the base plugin.
`after_setup`(dataset_size, mini_batch_size, ...)	Initialize different schedulers, including anneal target for AnnealOnPlateau, batch_growth_annealing, loading schedulers.
`after_training_batch`(optimizer_was_run, **kwargs)	Do the scheduler step if one-cycle or linear decay.
`attach_to`(pluggable)	Attach this plugin to a Pluggable.
`before_training_epoch`(**kwargs)	Load state for anneal_with_restarts, batch_growth_annealing, logic for early stopping.
`detach`()	Detach a plugin from the Pluggable it is attached to.
`get_state`()
`hook`([first_arg])	Convience function for BasePlugin.mark_func_as_hook).
`mark_func_as_hook`(func, *events)	Mark method as a hook triggered by the Pluggable.
`store_learning_rate`()

Attributes

`attach_to_all_processes`	If set, the plugin will be attached to all processes when distributed, not just the main process.
`corpus`
`model`
`pluggable`
`trainer`

after_setup(dataset_size, mini_batch_size, max_epochs, **kwargs)View on GitHub #: Initialize different schedulers, including anneal target for AnnealOnPlateau, batch_growth_annealing, loading schedulers.

before_training_epoch(**kwargs)View on GitHub #: Load state for anneal_with_restarts, batch_growth_annealing, logic for early stopping.

after_training_batch(optimizer_was_run, **kwargs)View on GitHub #: Do the scheduler step if one-cycle or linear decay.