Recurrent Neural Network (LSTM) Stock Model

Description

This model uses stock data and twitter sentiment to generate a prediction of future market trends. It gathers sentiment from scraping the Twitter website. Unfortunately, one cannot use the Twitter API as tweets are only available in a window spanning back two weeks. This tool solves this problem.

General process is as follows:

1. Get stock data from Quandl (day-to-day data)

2. Get stock twitter sentiment for each day

3. Join data

4. Feed it into a Long-term Short-term (LSTM) Neural Network

5. Graph past/future predictions

Directory Structure

Version1__base: Contains a basic implementation. No twitter sentiment analysis.

Version2__twitter_sentiment: Contains full functionality. Makes predictions with sentiment analysis included.

got/got3: Get Old Tweets. This repository can be found here.

img: Contains demo graphs.

misc: For implementations I might use later. For brainstorming.

model.py: The Recurrent Neural Network I am using to train and make predictions.

read_tickers.py: Python script to get all of the tickers in NASDAQ.

scratch.py: Playground script.

Performance Examples