Architecture Augmentation for Performance Predictor via Graph Isomorphism

This repository is for the paper "Architecture Augmentation for Performance Predictor via Graph Isomorphism" which is accepted by IEEE Transactions on Cybernetics.

Prerequisites

The codes have been tested on Python 3.6.

• nasbench (see https://github.com/google-research/nasbench)
• nas_201_api (see https://github.com/D-X-Y/NAS-Bench-201)
• tensorflow (==1.15.0)
• scikit-learn
• matplotlib
• scipy

Reproducing Results

NAS-Bench-101

1. Download the NAS-Bench-101 dataset and put in the data directory in the root folder of this project.
2. Run python nasbench101_pp.py to get the ktau of various performance predictors using training data with different proportions.

NAS-Bench-201

1. Download the NAS-Bench-201 dataset and put in the data directory in the root folder of this project.
2. Run python nasbench201_pp.py to get the ktau of various performance predictors using training data with different proportions.