Go to file
Jack Turner b74255e1f3 v2
2021-02-26 16:12:51 +00:00
autodl v2 2021-02-26 16:12:51 +00:00
config_utils v2 2021-02-26 16:12:51 +00:00
datasets v2 2021-02-26 16:12:51 +00:00
models v2 2021-02-26 16:12:51 +00:00
nas_101_api v2 2021-02-26 16:12:51 +00:00
nas_201_api v2 2021-02-26 16:12:51 +00:00
pycls v2 2021-02-26 16:12:51 +00:00
.gitignore Remove call to .eval(), update results in README, ignore .t7 files 2020-07-03 12:19:44 +01:00
env.yml v2 2021-02-26 16:12:51 +00:00
nasspace.py v2 2021-02-26 16:12:51 +00:00
plot_scores.py v2 2021-02-26 16:12:51 +00:00
README.md v2 2021-02-26 16:12:51 +00:00
score_networks.py v2 2021-02-26 16:12:51 +00:00
scorehook.sh v2 2021-02-26 16:12:51 +00:00
scores.py v2 2021-02-26 16:12:51 +00:00
search.py v2 2021-02-26 16:12:51 +00:00
utils.py v2 2021-02-26 16:12:51 +00:00

Neural Architecture Search Without Training

⚠️ Note: this repository has been updated to reflect the second version of the paper to appear on arXiv 1 March. :warning

Usage

Create a conda environment using the env.yml file

conda env create -f env.yml

Activate the environment and follow the instructions to install

Install nasbench (see https://github.com/google-research/nasbench)

Download the NDS data from https://github.com/facebookresearch/nds and place the json files in naswot-codebase/nds_data/ Download the NASbench101 data (see https://github.com/google-research/nasbench) Download the NASbench201 data (see https://github.com/D-X-Y/NAS-Bench-201)

Reproduce all of the results by running

./scorehook.sh

The code is licensed under the MIT licence.

Citing us

If you use or build on our work, please consider citing us:

@misc{mellor2020neural,
    title={Neural Architecture Search without Training},
    author={Joseph Mellor and Jack Turner and Amos Storkey and Elliot J. Crowley},
    year={2020},
    eprint={2006.04647},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}