Stock Trading Bot using Deep Reinforcement Learning (Deep Q-learning), Keras and TensorFlow. Reinforcement Learning is a type of machine learning that allows us to create intelligent agents that learn from the environment by interacting with it, it learns by trial and error. After each action, the agent receives the feedback in the form of reward or penalty. The feedback consists of the reward and next state of the environment. The reward is usually defined by a human. We can define reward as the profit from selling the stock bought at the original starting point.

Trading GOOGL, 2018 with a profit of $517.44

You can either use the Historical Financial data already present under data/ or download your own from Yahoo! Finance for training.

In order to train the model, you will need to install the required python packages:

pip install -r requirements.txt

Now you can open up a terminal and start training the agent:

python train.py --train-stock data/GOOGL.csv --val-stock data/GOOGL_2018.csv

Once you're done training, run the evaluation script and let the agent make stock decisions:

python evaluate.py --eval-data data/AAPL_2018.csv --window-size 10 --model-name model_GOOGL_10 --debug

Now you are all set up!

• Deep Q-Learning with Keras and Gym
• Playing Atari with Deep Reinforcement Learning
• Human Level Control Through Deep Reinforcement Learning

• Python
• TensorFlow
• NumPy

The project is a modification/impovement over edwardhdlu/q-trader.