/kekas

Just another DL library

Primary LanguageJupyter NotebookMIT LicenseMIT

Kekas

Kek it easy.

Kekas is a simple tool for training neural networks on Pytorch.

I tried to keep it as simple as possible.

Installation

pip install kekas

Mixed Precision requirements

Kekas use https://github.com/NVIDIA/apex library for mixed precision training, so follow the installation instructions from its repo.

Features

  • Awesome name
  • Mixed Precision (FP16)
  • Learning Rate Finder
  • One Cycle policy
  • Best checkpoints saving
  • Early stopping
  • TTA
  • Freeze / unfreeze
  • Easy customization

Quick start guide

I don't beieve in quick start guides, I think that they create more questions than answers.

Instead, I've created a detailed Tutorial notebook. Read it.

Contribution guide

Just contribute something good, and don't contribute anything bad.

TODO

  • AdamW
  • FP16 support
  • Discriminative learning rates
  • Docs :kekeke:

Citing

If you find this library useful for your research, please consider citing:

@misc{aleksandr belskikh_2019,
    author      = {Aleksandr Belskikh},
    title       = {{kekas: Just another DL library}},
    month       = jul,
    year        = 2019,
    doi         = {10.5281/zenodo.2577861},
    version     = {0.1.18},
    publisher   = {Zenodo}, 
    url         = {https://doi.org/10.5281/zenodo.2577861},
    }

DOI