Pytorch Learning Rate Finder

This package can be used to find optimal learning rate.

The package includes LearningRateFinder class which implements the fit, find_optimal_lr method . The fit method is used to find optimal learning rate within a range (optional)

Installation

To install with pip run the following command

pip install pytorch-lr-finder

Dependencies

This package requires the following to be installed:

• Python 3.6 or higher
• Pytorch
• Numpy
• Pandas
• Matplotlib

Instruction for usage

LearningRateFinder takes instantiated pytorch model (nn.module), criterion and optimizer (torch.optim).

The fit method requires a dataloader (torch.utils.data.DataLoader), you can optionally include the number of steps, the starting and ending learning rate. The plot function can be used to visualize the results in a plot. Please follow the example below for reference.

lrf = LearningRateFinder(model, criterion, optimizer)
lrf.fit(train_loader)
lrf.plot()

Example