This is Python 3.6 code for implementing a Deep Q neural networks for algorithmic trading within the cryptocurrency/stock market.It's implementation of Q-learning applied to (short-term) trading. The model uses n-day windows of High prices (Given that is the quantum algorithm decision for choosing the feature at the time) (Time,Open,High,Low,Close,Volume) to determine if the best action to take at a given time is to buy, sell or hold.
Results of the examples included in datasets feature rewards, test scores, and actions for the state of a given stock price.
75 = window(# of days)
200 = episodes
Step 1:
portfolio_management>python fulldataimport.py
(if in a rush remove the stock import feed)
step1 will extract the text files and real time intraday cryptocurrency data into the data format suitable for the machine learning simulation environment for the Deep Q Network.
2:
--> OPEN "01-feature-selection.ipynb". This is the quantum feature selection algorithm
--> OPEN "Linear_Reg__.ipynb" -- this is the benchmark model for analyzing and predicted time segments between 15-5 minutes.
portfolio_management > python train2.py crypto/crypto_portfolio/15m/bitfinex_ethusd 75 200
portfolio_management > python train2.py crypto/crypto_portfolio/5m/bitfinex_ethusd 25 200
** minimum 200 episodes for results
** TRAINING MODELS CAN BE RAN AT THE SAME TIME BUT HIGHLY UNRECOMMENDED.
portfolio_management > python evaluate2.py crypto/crypto_portfolio/15m/bitfinex_ethusd agent2/eth/model_ep-200
portfolio_management > python evaluate2.py crypto/crypto_portfolio/5m/bitfinex_ethusd agent2/eth/model_ep-200
Test simulation results against benchmark regressor while analyzing data within 15 minute -5 minute windows. LINEAR REGRESSOR IPYNB NOTEBOOK WILL BE THE BENCHMARK MODEL USING LINEAR REGRESSION AND TIME SERIES CROSS VALIDATION.. AND COMPARE IT TO THE BACKTEST SIMULATION RESULTS. Add in psychology behind the trades. Add indicators (news (sentiment analysis, Boilinger bands, Volume Indicator)).
OVERALL model_ep-200 WILL BE USED VIA COINBASE API KEYS FOR ALGORITHMIC TRADING