Skip to content

Attempting to use a Constitutional Neural Network to make daily high-low-close predictions from previous daily data.

Notifications You must be signed in to change notification settings

joe-habel/StockCNN

Repository files navigation

StockCNN

Attempting to use a CNN to make daily high-low-close predictions from previous intra-day data.

This is a work in progress to scrape intra-day ticker data and use that as a continuous 1D-signal, and utilize a CNN to attempt to predict the following day's HLC.

symbols_getter.py scrapes all of the individual tickers.

historical_scraper.py scrapes the entire availble history of minute by minute data from all the the tickers from symbols_getter.py

preprocess_json.py parses out the json format from historical_scraper.py into a csv format.

make_training.py parses the csv data into a labeled training data.

CNNreg.py uses the training data to train a CNN, with the labeled previous day (390,1) array with the following day's HLC.

predictions.py uses the CNN to make predictions on the next available day in the data.

About

Attempting to use a Constitutional Neural Network to make daily high-low-close predictions from previous daily data.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages