/Transfer-Learning-for-Reinforcement-Learning

a multi-threaded AlphaGo-style agent to iteratively learn from self-play data using Monte-Carlo Tree Search algorithm

Primary LanguagePythonMIT LicenseMIT

SuperGo For CZ4041

Environment:

Ubuntu 18.04
Python 3.5
CUDA 10.1
Local Mongo DB Server

Python Library Required:

pachi_pi
numpy
timeit
pickle
pytorch 1.4
pymongo

Training Configuration:

LR: 0.01
BoardSize 99
Data Batch Size: 4000 steps
MiniBatch: 64
Iteration for one generation: 12MiniBatch 15*MiniBatch

To start Traning from zero:

python main.py

To start training from break point

pythoin main.py --folder your_folder --verion version_number