rl-rs-trader

Additional resources

• Pdf report of project details (Final-Project-Report.pdf in repo)
• Slide show of project overview (Final-Slides in repo)
• Link to slide presentation - https://youtu.be/LddPK043qfw

Required packages (should be obtainable from pip-install)

• torch
• numpy
• matplotlib
• seaborn
• pandas

You can run this code and running the rl-mercher.py file with the following parameters

• <agent_type> either TD or DQN
• <model_load> names are only required if continuing to train an existing model that needs to be loaded
• <model_targets> only needed if doing dqn because they require the two model files

if training.

python rl-mercher.py train <model_save_name> <agent_type> <model_load_name> <model_target_load_name>

if evaluating.

python rl-mercher.py eval <eval_model_name> <agent_type> <model_eval_name> <model_target_eval_name>

Explanation of files

rl_mercher.py

main file to run and train models runnable via command line arguments as mentioned above other paramters configurable by editing the file iteself (random-decay, minimum_randomness)

looper_helpers.py

various helper methods using in rl_mercher to clean up main file includes the functions handling the learning episode loop and the model evaluation loop

ge_env.py

Open-Gym enviornment to simulate trading items on the grand exchange

td_agent.py

Temporal Difference Agent

dqn_agent.py

Deep Q-Learning Network Agent

item_data.csv

Data file with the historic price/volume information of different game items