elbowz / ml-time-series-prediction

Machine Learning: Time series prediction and strategies on stock markets

Time series prediction and strategies on stock markets

more detailed information in ITALIAN: REPORT.MD

Utilized libs

• keras: neural-network library running on TensorFlow
• sklearn: encoding, metrics and classificator models
• pandas: load and manipulation of DataSet
• numpy, collections: advanced data structures
• talib: market indicators (Technical analysis)
• alpha_vantage: online data market source
• json, hashlib, typing, argparse, matplotlib, os, etc...

Usage

main.py [-h] [--in-memory-training] [--symbol-data-overwrite] [--force-train] CFG_FILENAME    

• CFG_FILENAME: path_string - path to config file
• --in-memory-training, -m: TrainingSet manipulation in memory (not suggested). Default False
• --symbol-data-overwrite, -d: overwrite the DataSet (historical data). Default False
• --force-train, -f: force retraining LSTM

nota: major part of configuration is made by config file (see /cfg directory)

Credits

• LSTM Neural Network for Time Series Prediction by Jakob Aungiers
• Alpha Vantage API by Romel Torres
• TA-Lib by John Benediktsson