Compiler Optimizing via Deep Reinforcement Learning (COREL)
Overview
This work aims to employee Deep Reinforcement Learning to solve Compiler Optimizing.
Currently Implemented
- AlphaCompile with MCTS
Roadmap (To be implemented in order)
- PPO for Compiler Optmization
- Genetic Compiler
Setup
To run this model clone the repo, create a new virtualenv, install requirements from requirements.txt, enjoy!
Running a model
Run any model as follows: python -m source.<ModelToRun>
Visualizing the Results
This project currently uses MLFlow, to startup the dashboard, call mlflow ui.
Generating Documentation
Documentation for the entire package can be generated by executing make from withing the docs/ directory. To view the resulting docs locally, simply open index.html in your favorite browser.
Release Notes
Version 1.0
- Switched from Keras to PyTorch
- Discarded all non-RL models
- Implemented AlphaCompile, based on AlphaZero architecture.