This is a small architecture and training pipeline using a CNN and integrating Random Fourier Features for making predictions of the stock market.
Maintain a Pandas pickle over Stock market data samples from Polygon. Daily.
Maybe there was originally an effort to cache this into an Elastic instance?
Train a set of models over a set of ticker symbols; each model specializes in predicting price over the specified symbols.
Thin wrapper over TR-1007-PFC
Trains a model and writes it to disk.
Augments data with Random-Fourier Features of specified geometry using a ThreadPool to accelerate performance.
Contains commented-out functionality for performing backtesting using a naive strategy, to be run after performing batch training.