BaseFinetuning

class pytorch_lightning.callbacks.BaseFinetuning

Bases: pytorch_lightning.callbacks.base.Callback

This class implements the base logic for writing your own Finetuning Callback.

Override freeze_before_training and finetune_function methods with your own logic.

freeze_before_training: This method is called before configure_optimizers

and should be used to freeze any modules parameters.

finetune_function: This method is called on every train epoch start and should be used to

unfreeze any parameters. Those parameters needs to be added in a new param_group within the optimizer.

Note

Make sure to filter the parameters based on requires_grad.

Example:

class MyModel(LightningModule)

    ...

    def configure_optimizer(self):
        # Make sure to filter the parameters based on `requires_grad`
        return Adam(filter(lambda p: p.requires_grad, self.parameters))

class FeatureExtractorFreezeUnfreeze(BaseFinetuning):

    def __init__(self, unfreeze_at_epoch=10)
        self._unfreeze_at_epoch = unfreeze_at_epoch

    def freeze_before_training(self, pl_module):
        # freeze any module you want
        # Here, we are freezing ``feature_extractor``
        self.freeze(pl_module.feature_extractor)

    def finetune_function(self, pl_module, current_epoch, optimizer, optimizer_idx):
        # When `current_epoch` is 10, feature_extractor will start training.
        if current_epoch == self._unfreeze_at_epoch:
            self.unfreeze_and_add_param_group(
                modules=pl_module.feature_extractor,
                optimizer=optimizer,
                train_bn=True,
            )

static filter_on_optimizer(optimizer, params)

This function is used to exclude any parameter which already exists in this optimizer

Parameters

• optimizer (Optimizer) – Optimizer used for parameter exclusion

• params (Iterable) – Iterable of parameters used to check against the provided optimizer

Return type

List

Returns

List of parameters not contained in this optimizer param groups

static filter_params(modules, train_bn=True, requires_grad=True)

Yields the requires_grad parameters of a given module or list of modules.

Parameters

• modules (Union[Module, Iterable[Union[Module, Iterable]]]) – A given module or an iterable of modules

• train_bn (bool) – Whether to train BatchNorm module

• requires_grad (bool) – Whether to create a generator for trainable or non-trainable parameters.

Return type

Generator

Returns

Generator

finetune_function(pl_module, epoch, optimizer, opt_idx)

Override to add your unfreeze logic

static flatten_modules(modules)

This function is used to flatten a module or an iterable of modules into a list of its modules.

Parameters

modules (Union[Module, Iterable[Union[Module, Iterable]]]) – A given module or an iterable of modules

Return type

List[Module]

Returns

List of modules

static freeze(modules, train_bn=True)

Freezes the parameters of the provided modules

Parameters

• modules (Union[Module, Iterable[Union[Module, Iterable]]]) – A given module or an iterable of modules

• train_bn (bool) – If True, leave the BatchNorm layers in training mode

Return type

None

Returns

None

freeze_before_training(pl_module)

Override to add your freeze logic

static make_trainable(modules)

Unfreezes the parameters of the provided modules

Parameters

modules (Union[Module, Iterable[Union[Module, Iterable]]]) – A given module or an iterable of modules

Return type

None

on_before_accelerator_backend_setup(trainer, pl_module)

Called before accelerator is being setup

on_train_epoch_start(trainer, pl_module)

Called when the epoch begins.

static unfreeze_and_add_param_group(modules, optimizer, lr=None, initial_denom_lr=10.0, train_bn=True)

Unfreezes a module and adds its parameters to an optimizer.

Parameters

• modules (Union[Module, Iterable[Union[Module, Iterable]]]) – A module or iterable of modules to unfreeze. Their parameters will be added to an optimizer as a new param group.

• optimizer (Optimizer) – The provided optimizer will receive new parameters and will add them to add_param_group

• lr (Optional[float]) – Learning rate for the new param group.

• initial_denom_lr (float) – If no lr is provided, the learning from the first param group will be used and divided by initial_denom_lr.

• train_bn (bool) – Whether to train the BatchNormalization layers.

Return type

None

Returns

None