ashawkey / hawtorch

pytorch extensions for code reuse

Hawtorch

This is a framework extension for Pytorch, aiming at code re-use, visualization and neat logging.

The codes are simple to read, and this package encourages users to read and understand the source codes when they encountered problems.

All you need to do is to write a pytorch model, a data provider and a main program to drive them.

Examples can be found under the corresponding folders.

Structure

• trainer

  A easy-to-modify general Trainer class for any model and metrics.

  • auto save, load checkpoints.
  • tqdm , tensorboardX integration

• io

  Simplified input & output API, and a logger class for creating logs while training.

  • beautiful and clear logging style.

• utils

  • DelayedKeyboardInterrupt

    This is used to avoid saving incomplete checkpoints.

  • EmailSender

    This can be used to send you an email reporting the reports and logs when the training process is ended.

• nn

  Custom Neural Networks, Functionals and Losses as in Pytorch.

• optim

  Custom optimizers.

• vision

  Simplified API for Image-specific operations.

  • plot image by filenames
  • 3D point cloud / graph

• metrics

  Evaluation metrics for different tasks.