OpenAI gym to play with stock market data

1. Download stock data in comma-separated CSV format, following fields are required 'Date', 'Time', 'Open', 'High', 'Low', 'Close', 'Volume' into stocks/ directory within this git sources, each separate equity datafile should have .csv extension
2. Install gym-stocks

pip install --user -e .

3. Run example

import gym
import gym_stocks
env = gym.make('Stocks-v0')
print env.reset()

4. Initial (reset) conditions You have 1000000 units of money and zero equity. Opeartion comission is 0.1%, there is no inflation (will be added if needed), i.e. negative reward per HOLD action.

gym-stocks opens one random csv file from stocks directory at every reset() call and yields one line per step. No normaization is being performed.

Every buy/sell action uses previous close price, i.e. not the price returned by the step() call, comission is being applied per each of these actions. Only 1 stock is being processed in these steps, i.e. if you select BUY, and the price is 50, only one stock will be bought (you will pay 50*(1+0.1/100)) even if you have more money.

If you do not have enough money to perform BUY action or you do not have equity to perform SELL action, nothing happens.

Portfolio cost equals to the sum of the money and equity times the closing price. Reward is a difference between portfolio cost at the current and previous steps, reward calculation is being performed after money and equity have been updated (with the appropriate comission).