LearningRateFinder¶

class pytorch_lightning.callbacks.LearningRateFinder(min_lr=1e-08, max_lr=1, num_training_steps=100, mode='exponential', early_stop_threshold=4.0, update_attr=False)[source]¶

Bases: pytorch_lightning.callbacks.callback.Callback

The LearningRateFinder callback enables the user to do a range test of good initial learning rates, to reduce the amount of guesswork in picking a good starting learning rate.

Parameters:

min_lr¶ (float) – Minimum learning rate to investigate
max_lr¶ (float) – Maximum learning rate to investigate
num_training_steps¶ (int) – Number of learning rates to test
mode¶ (str) –
Search strategy to update learning rate after each batch:
- 'exponential' (default): Increases the learning rate exponentially.
- 'linear': Increases the learning rate linearly.
early_stop_threshold¶ (Optional[float]) – Threshold for stopping the search. If the loss at any point is larger than early_stop_threshold*best_loss then the search is stopped. To disable, set to None.
update_attr¶ (bool) – Whether to update the learning rate attribute or not.

Example:

# Customize LearningRateFinder callback to run at different epochs.
# This feature is useful while fine-tuning models.
from pytorch_lightning.callbacks import LearningRateFinder


class FineTuneLearningRateFinder(LearningRateFinder):
    def __init__(self, milestones, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.milestones = milestones

    def on_fit_start(self, *args, **kwargs):
        return

    def on_train_epoch_start(self, trainer, pl_module):
        if trainer.current_epoch in self.milestones or trainer.current_epoch == 0:
            self.lr_find(trainer, pl_module)


trainer = Trainer(callbacks=[FineTuneLearningRateFinder(milestones=(5, 10))])
trainer.fit(...)

Raises:: MisconfigurationException – If learning rate/lr in model or model.hparams isn’t overridden when auto_lr_find=True, or if you are using more than one optimizer.

on_fit_start(trainer, pl_module)[source]¶

Called when fit begins.

Return type:: None