This is the code repository of "A Modular Framework for RL Optimal Execution".
We recommend running the experiments via the Dockerfile
.
Alternatively, one can manually install all the necessary libraries in a virtual environment via
pip install -r requirements.txt
The train_{algo}.py
files are the entrypoints of the experiments. Training and evaluations can be carried out modifying their main()
functions. Adding new agents and training on different periods (if the data is provided) can be done via modifying the config
dict.
src/core/data/historical:datafeed.py
contains the implementation of the DataFeed
class.
src/core/environment/limit_orders_setup
contains the implementations of the Execution Algo
, Broker
classes as well as the gym
environment.
src/tests
contains the implementation of the multiple UnitTests applied to the environment.
The plot_schedule
method of the Execution Algo
class can reproduce Figures 5 & 6 of the paper.