shikharbhardwaj/tradegame

An algorithmic trader using Reinforcement Learning.

tradegame

An Algorithmic Trader using Reinforcement learning, Financial Trading as a Game: A Deep Reinforcement Learning Approach.

Dependencies

See requirements.txt.

Training and evaluation

To train a new model, configure the training using the JSON config file, a sample is provided in sample.json

{
    "data_location": "/data/tradegame_data/sampled_data_15T",
    "pairs": ["AUDJPY", "AUDNZD", "AUDUSD", "CADJPY", "CHFJPY", "EURCHF",
        "EURGBP", "EURJPY", "EURUSD", "GBPJPY", "GBPUSD", "NZDUSD", "USDCAD",
        "USDCHF", "USDJPY"],
    "trade_pair": "EURUSD",
    "begin_year": 2012,
    "end_year": 2017,
    "start_cash": 200000,
    "trade_size": 100000
}

Execute the training script

$ python train.py config.json

To evaluate a trained model update the JSON config file for testing, a sample is provided in test.json

{
    "data_location": "D:\\tradegame_data\\sampled_data_15T",
    "model_location": "E:\\dev\\tradegame\\src\\models\\flat_state_exp\\eval_model2.h5",
    "pairs": ["AUDJPY", "AUDNZD", "AUDUSD", "CADJPY", "CHFJPY", "EURCHF",
        "EURGBP", "EURJPY", "EURUSD", "GBPJPY", "GBPUSD", "NZDUSD", "USDCAD",
        "USDCHF", "USDJPY"],
    "trade_pair": "EURUSD",
    "begin_year": 2017,
    "end_year": 2018,
    "start_cash": 200000,
    "trade_size": 100000
}

Execute the evaluation script

$ python train.py test.json