Using Reinforcement Learning agents as Algorithmic Traders
This repository contains the source code for my project report, for the course CS 885 (Reinforcement Learning), taught by Professor Pascal Poupart.
To install and run, use the file environment.lock.yaml and conda.
The notebooks contain (very messy) code used to experiment, and develop, the source code in the trading/ folder.
In particular, notebooks/experiments.ipynb generated most of the illustrations in the plots/ folder.
This part of data pipeline is intentionally not automatic.
To acquire the Reddit comments dataset from Christopher Lambert’s Kaggle release:
> wget "https://www.kaggle.com/theriley106/wallstreetbetscomments?select=wsbData.json"
You will need to download the Common Crawl word vectors. See here for how that can be done.
Store the unzipped file in the top-level of this cloned repository.
Run the script trading/wsb_pipeline.py to create a database of comment vectors. These embeddings are used to train ModelBasedAgent objects,
as they are processed through TradingWithRedditEnv.