NASLib is a Neural Architecture Search (NAS) library for facilitating NAS research for the community by providing interfaces to several state-of-the-art NAS search spaces and optimizers.
Use
python naslib/benchmarks/statistics/runner.py --config-file naslib/benchmarks/statistics/statistics_config.yaml
or
bash naslib/benchmarks/statistics/run.sh
to compute statistics such as mean, std, max, min val accs, autocorrelation, avg neighborhood size, etc.
Works for all search spaces except nas-bench-asr
Use
python naslib/benchmarks/statistics/runner.py --config-file naslib/benchmarks/statistics/statistics_config.yamlorbash naslib/benchmarks/statistics/run.shto compute statistics such as mean, std, max, min val accs, autocorrelation, avg neighborhood size, etc. Works for all search spaces except nas-bench-asr