Stock Trading Agent

Directory Structure

stock-trading-agent
│   main.py                 # main script for training and testing
│   StocksEnv.py            # Gym environment for stock trading
│   StocksData.py           # util for processing stock data
│   requirements.txt        # list of Python dependencies
│   Dockerfile              # Docker image reference
│   README.md               # this file
│   
└───datasets                # input stock data for training and testing
│   │   appl_test.csv
│   │   appl_train.csv
│   │   googl_test.csv
│   │   googl_train.csv
│   │   ...
│   
└───models                  # output model files from training
│   │   appl_model.zip
│   │   googl_model.zip
│   │   ...
│
└───results                 # output results from testing
│   │   appl_x_y.csv        # x: net reward, y: timestamp
│   │   googl_x_y.csv
│   │   ...
│
└───tensorboard             # output tensorboard logs while training
    │   TRPO_1
    │   TRPO_2
    │   ...

Quickstart

Note: the following instructions have been tested to work on macOS 10.15.4 with Docker Desktop installed and running

Obtain Docker Image

To build the image, run the following command from the root directory...

docker build -t stock-trading-agent:latest .

Alternatively, a pre-built image can be downloaded with...

docker pull raz4/stock-trading-agent:latest

Run Training and Testing

Note: In order to use input files on and output files to the "host" filesystem, the respective volumes must be mounted to the running container. The "-v" options in the following command define the mapping between the host and container directories.

docker run \
-v "$(pwd)"/models:/app/models \
-v "$(pwd)"/datasets:/app/datasets \
-v "$(pwd)"/results:/app/results \
-v "$(pwd)"/tensorboard:/app/tensorboard \
stock-trading-agent:latest \
--training_data=/app/datasets/appl_train.csv --training_timesteps=20000 --model_file=./models/appl_model --testing_data=/app/datasets/appl_test.csv