Savant is an artificial neural network system for analyzing and predicting changes in the stock market, targeted at predicting shifts in the S&P 500 (as modeled by the SPY exchange-traded-fund).
Objective: Given historical market data, predict the future price of the SPY ETF (which models the S&P 500 index)
Possible Indicators (to be modified later pending cross-validation): + SPY (last 30 days) * An ETF that models the S&P 500 index + USO (last 30 days) * An ETF that tracks the performance of the price of crude oil + SHY (last 30 days) * An ETF that tracks an index of U.S. Treasury bonds with maturities between one and three years
Note: We have decided to use ETFs because, as they are traded like securities, historical pricing data is readily available for ETFs, simplifying the data-acquisition component of the system pipeline
Machine Learning Technique: Artificial Neural Network
Technology Stack: Scipy, Numpy, Pandas, & matplotlib (iPython used for development)
spy.pycontains the code for retrieving historical price data from the Yahoo Finance API.tailor.pycontains the code for extracting relevant data from the data retrieved byretriever.py& constructing feature vectors and matrices.soldier.pycontains the actual neural network code for running forward- and back-propagation.tinker.pycontains the functions for training the network and making new predictions.circus.pycontains the central script for getting data, training the network, cross-validating, and so-on.
- Although many comments refer to objects as "matrices", those objects are actually
numpy.ndarrays. The phrase "matrix" is used only in the mathematical sense.
Copyright Charles Dawson 2015. Code is posted here for historical & reference purposes.