Reinforcement Learning Enhanced Quantum-inspired Algorithm for Combinatorial Optimization

This is the official implementation of the paper "Reinforcement Learning Enhanced Quantum-inspired Algorithm for Combinatorial Optimization"

Setup the environment

conda env create --name sim -f environment.yml
conda activate sim
pip install git+https://github.com/BeloborodovDS/baselines.git
conda install -c conda-forge tensorflow

Run experiments:

• make train_ref: pre-train the agent an random problems (R3, FILM)
• make train_R2: pre-train the agent an random problems (R2, FILM)
• make train_nofilm: pre-train the agent an random problems (R3, no FILM)
• make experiment_ref: train the agent on graphs G1-G10 (fine-tune, R3, FILM)
• make experiment_R2: train the agent on graphs G1-G10 (fine-tune, R2, FILM)
• make experiment_nofilm: train the agent on graphs G1-G10 (fine-tune, R3, no FILM)
• make experiment_scratch: train the agent on graphs G1-G10 (from scratch, R3, FILM)
• make experiment_R2_scratch: train the agent on graphs G1-G10 (from scratch, R2, FILM)
• make experiment_nofilm_scratch: train the agent on graphs G1-G10 (from scratch, R3, no FILM)

See plot.ipynb for plots and tables from the paper.

Data sources

Gset (data/G{i}.txt): dataset link

Best cuts (data/gbench.txt): paper link