README.md

This code is implemented according to the paper "Cost-Aware Graph Generation: A Deep Bayesian Optimization Approach", accepted by AAAI 2021. Cost-Aware Graph Generation (CAGG) can generate optimal graphs at as low cost as possible. We apply it to two challenging real-world problems, i.e., molecular discovery and neural architecture search, to rigorously evaluate its effectiveness and applicability.

An illustrative task 1: Molecular Discovery on QM9 data

An illustrative task 2: Cell-based Neural Archtecture Search on CIFAR100 data

Folder description:

CAGG-Molecular-Discovery: Molecular Discovery, including two molecular properties, i.e., 5*QED-SA and logP-SA
CAGG-NAS: Neural Architecture Search (NAS), including cell-based NAS and multi-branch NAS

Please go to their respective folders for details.

This code has been tested on the environment with MacOS and Python3.