An assumption-free microscopic stock market model built upon multi-agent reinforcement learning.
This is a final project for the course CS285 Deep Reinforcement Learning, Decision Making, and Control at UC Berkeley. Please see our final report for more details about this project.
Use the package manager pip to install the dependencies locally.
git clone -b master --depth 1 https://github.com/ChocolateDave/stock_market.git
cd stock_market & pip install -r requirements.txt & pip install -e .
We provide three scripts for running our codes on different settings.
-
If you would like to explore agents trained on different learning rates, please run with
bash stock_market/scripts/run_stock_market_diff_lr $GPU_ID
-
If you would like to investigate agents trained on different budget discount over time, please run with
bash stock_market/scripts/run_stock_market_budget_discount $GPU_ID
-
If you would like to investigate agents trained on different worth of stock, please run with
bash stock_market/scripts/run_stock_market_worth_of_stocks.sh $GPU_ID
-
Or you can run the base training script for more flexibility
python stock_market/train.py <$ARGUMENTS>
If you use this source code, please cite it using bibtex as below.
@misc{maverick2022,
author = {Maverick, Zhang and Juanwu Lu},
title = {Multi-Agent Reinforcement Learning for Assumption-Free Stock Market Modeling},
year = {2022},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/ChocolateDave/stock_market}},
commit = {2b675bac077bc695048ce0072f254de25c898050}
}
This project is licensed under the BSD 3-Clause License