DRL-trading-bot

Deep Reinforcement learning trading bot

This project provide tools and utils to train and test Deep Reinforcement Learning Trading Agent on Crypto Currency market

Data Fetcher

Python script to fetch the dataset from the Binance Exchange.

./fetch_data.py binance --key secrets.json --period 1h --symbol "BTCUSDT" --start-date "2019-01-01" --end-date "2021-01-01" --output dataset/

• --key: Secret key (API key) for the binance exchange.
• --period: Period of data (EX: '1m' '1h' '1d' '1m').
• --symbol: Crypto Currency pair symbol (EX: BTCUSDT).
• --start-date: Data staring date.
• --end-date: Data ending date.
• --output: Output folder to save dataset to.

Trading Environment

• Data Provider which is responsible for providing data and observations for the environment.
• Reward which is responsible for calculating the value of the reward.
• Trading Strategy which is responsible for executing the actions provided by the agent.
• Rendering component which is responsible for rendering and providing a visual representation of the trading environment.

Training/Testing CLI

Training

./optimize.py ray_train -a PPO -r sortino -data dataset/binance-BTCUSDT-1h-2019-01__2021-01.csv --add-indicators

• -a: Algorithm used to train the agent (EX: PPO, A2C ..).
• -r: Reward function used (EX: simple_profit, sharp, sortino).
• -data Dataset used to train.
• --add_indicators: This flag when set technical indicators are added to the observation space.
• -lstm: This flag wraps agent network with lstm layer.

Testing

./optimize.py ray_test -a PPO -r simple_profit -data dataset/binance-BTCUSDT-1h-2019-01__2021-01.csv --add-indicators -id binance-BTCUSDT-1h-lite_0_2022-05-13_17-53-55 --episodes 3 -ns 160 --render

• -a: Algorithm used to train the agent (EX: PPO, A2C ..).
• -r: Reward function used (EX: simple_profit, sharp, sortino).
• -data Dataset used to train.
• --add_indicators: This flag when set technical indicators are added to the observation space.
• -lstm: This flag wraps agent network with lstm layer.
• -id: Id of the model we want to test.
• --episodes: Number of episodes to test on.
• -ns: List checkpoints we want to test.
• --render: This flag trigger environment rendering.

Backtesting platform

To deploy production version

cd client && docker-compose up --build

To deploy development version

cd client

# run api
docker-compose -f docker-compose-dev.yml up --build

# run front 
cd rl-trader-front 
yarn && yarn start