What it does This is a student/academic project that applies supervised learning to stock price forecasting. It pulls historical price data from Yahoo Finance, NASDAQ, and Google Finance, layers on Twitter sentiment analysis, then runs the kitchen sink of 2016-era techniques—LSTMs, RNNs, ARIMA, and classical regression—at the problem. The stated goal is portfolio construction through diversification, though the README is vague on how that actually happens.

The interesting bit The bibliography is doing serious heavy lifting. Three Stanford CS229 papers, a full methodology section, slides, a video, and a PDF report—this is coursework dressed for a conference. The Twitter sentiment integration is a nice touch for the era, back when “social signals” felt novel rather than noisy.

Key highlights

• Implements multiple approaches: LSTM/RNN neural networks, ARIMA time-series models, and scikit-learn regression baselines
• Pulls in external sentiment data from Twitter as a feature
• Includes a concept video and 30-slide deck (linked externally)
• References Theano, which dates the project roughly to the Obama administration
• Dataset lives on a public Google Drive link of uncertain longevity

Caveats

• Requires manual dataset download from a Google Drive link; no automated data pipeline
• Built on Theano, which is now defunct; expect dependency archaeology
• README describes methodology but gives almost no quantitative results or model comparisons
• Single entry point (regression_models.py) suggests the LSTM/ARIMA code may be scattered or require manual orchestration

Verdict Worth a skim if you’re writing a literature review on ML-for-finance circa 2015–2017, or if you need to explain to a manager why “just add LSTM” isn’t a trading strategy. Skip if you want production code, reproducible benchmarks, or anything that runs without a time machine.