UdonDa / DARTS_pytorch

Non official implementation of DARTS[Liu+, ICLR19] https://arxiv.org/abs/1806.09055

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

DARTS: Differentiable Architecture Search

This repository is a non-official implementation.

Introduction of DARTS

https://speakerdeck.com/udonda/darts-differentiable-architecture-search

Usage

  1. Run bash scripts/1_search.sh
  2. Set the output of training script to genotype parameters in scripts/2_augment.sh
  3. Run bash scripts/2_augment.sh

Results

CIFAR-10 Final validation acc Best validation acc
Original 97.17% 97.23%
Ours 96.39% 96.87%

Causion

  • My implementation can not handle multiple-gpu training.

Acknowledgements

@article{liu2018darts,
  title={DARTS: Differentiable Architecture Search},
  author={Liu, Hanxiao and Simonyan, Karen and Yang, Yiming},
  journal={arXiv preprint arXiv:1806.09055},
  year={2018}
}

About

Non official implementation of DARTS[Liu+, ICLR19] https://arxiv.org/abs/1806.09055


Languages

Language:Python 98.4%Language:Shell 1.6%