This repository contains an level 3 order book simulation for data from the GDAX API. The simulation is made so that it fit the abstract API of a OpenAI gym class.

The simulation of the order book is inpspired by .. but modified and cythonized.

A conda environment yaml exists.
conda env create -f environment.yml
To build cython
python setup.py build_ext --inplace

But other python dependecies exists
for OpenAI
Install everything
for Ray
Dependencies

The data can be downloaded by download.js, and then converted to the feather format by running "Save data as Feather.ipynb"

A smaller data set exists at
data

One class is currently developed, where the action space is the alpha and beta of a gamma distribution. So the buy and sell limit orders are placed according to a gamma distribution, see MarketBetaEnv

The policy is optimized using PPO.

The environment can be rendered as a dash app. Under development.