dhruvrajgupta / nas

NASLib

NASLib is a Neural Architecture Search (NAS) library. Its purpose is to facilitate NAS research for the community by providing interfaces to several state-of-the-art NAS search spaces.

⚠️ This library is under construction and there is no official release yet. Feel free to play around and have a look but be aware that the APIs will be changed until we have a first release.

It is designed to be modular, extensible and easy to use.

Usage

search_space = SimpleCellSearchSpace()

optimizer = DARTSOptimizer(config)
optimizer.adapt_search_space(search_space)

trainer = Trainer(optimizer, config)
trainer.search()        # Search for an architecture
trainer.evaluate()      # Evaluate the best architecture

For an example file see demo.py.

For more extensive documentation see docs.

NASLib has been used to run an extensive comparison of 31 performance predictors. See the separate readme: predictors.md and our paper: How Powerful are Performance Predictors in Neural Architecture Search?

Installation

Clone and install.

If you plan to modify naslib consider adding the -e option for pip install.

git clone -b dllab22 https://github.com/automl/NASLib/
cd NASLib
conda create -n naslib_exercises python=3.7
conda activate naslib_exercises
pip install --upgrade pip setuptools wheel
pip install -e .
pip install gdown jupyter

To validate the installation, you can run tests:

cd tests
coverage run -m unittest discover

The test coverage can be seen with coverage report.

Download data

To download the tabular benchmark and setup the data folder run

source scripts/download_data.sh nb201 cifar10

Tutorial

Please refer to docs/naslib_tutorial.ipynb for instructions on the tutorial and exercises

Cite

If you use this code in your own work, please use the following bibtex entries:

@misc{naslib-2020, 
  title={NASLib: A Modular and Flexible Neural Architecture Search Library}, 
  author={Ruchte, Michael and Zela, Arber and Siems, Julien and Grabocka, Josif and Hutter, Frank}, 
  year={2020}, publisher={GitHub}, 
  howpublished={\url{https://github.com/automl/NASLib}} }
  
@article{white2021powerful,
  title={How Powerful are Performance Predictors in Neural Architecture Search?},
  author={White, Colin and Zela, Arber and Ru, Binxin and Liu, Yang and Hutter, Frank},
  journal={arXiv preprint arXiv:2104.01177},
  year={2021}
}