This repository contains our final project for cs230. It is a deep neural network (LSTM network implemented in Keras) that predicts the future price of Bitcoin (in USD) based on historic order books data on the Gemini cryptocurrency exchange.
After cloning this repository, you will need to get the gemini order book dataset. The dataset we used are order books from 10/08/2015 - 02/20/2018. You can buy this dataset at https://datashop.cboe.com/cryptocurrency-gemini-order-book-data
More details can be found in poster.jpg and report.pdf
The gemini orders book dataset comes as a bunch of zip files that you download from their FTP server
You first want to convert those into parquet files - use the scripts convert_to_lz4.py and to_parquet.py to do this
Then you can run preprocessing.ipynb to extract out a bunch of features from the order books
The training and testing procedure for the best model we found is in training_testing_FINAL_MODEL.ipynb
The models directory contains several pre-trained models that can be loaded via Keras
MIT