Free Stock‑prediction repo posted

Quant Science pointed followers on April 12 to a free GitHub stock-forecasting project built in Python, framing it as hands-on code for machine learning and deep learning work on equities. (x.com) The post linked to `borisbanushev/stockpredictionai`, a public GitHub repository that says it builds a “complete process” for predicting stock price movements. GitHub’s cached listing shows the repo had about 4,800 stars and 1,800 forks by 2025, a sign that it was already widely circulated before Quant Science resurfaced it. (github.com, github.com) At its core, stock prediction code tries to learn patterns from old market data and then estimate what comes next. This repository layers several inputs together, including price history, technical indicators, news sentiment from Bidirectional Encoder Representations from Transformers, Fourier transforms, autoregressive integrated moving average features, and feature ranking with XGBoost. (github.com, en.rattibha.com) The project description says the example uses Goldman Sachs as a sample stock but draws on data from 72 assets, separating training data from test data with a visible split in the notebook. It also says the workflow uses a generative adversarial network, with a long short-term memory model as generator and a convolutional neural network as discriminator. (en.rattibha.com, github.com) Quant Science’s interest fits its broader business. Its site sells algorithmic trading courses, and its GitHub profile lists three public repositories, with `sunday-quant-scientist` as the largest at about 1,700 stars and 358 forks. (quantscience.io, github.com, github.com) That context matters because free code repositories often serve two jobs at once: they are learning material for programmers and marketing funnels for training products. Quant Science’s website pitches a free newsletter, a five-day course, and a paid system for algorithmic trading. (quantscience.io, github.com) The repository itself also shows the limits of this corner of finance education. Multiple GitHub issues ask for the missing dataset or for code behind parts of the notebook, including the generative adversarial network and Bidirectional Encoder Representations from Transformers sections. (github.com, github.com) Critics on the project’s issue tracker have also questioned whether a stack that mixes autoregressive integrated moving average models, sentiment analysis, and generative adversarial networks can reliably predict stock prices at all. The repo’s own materials present it as an educational walkthrough of techniques rather than a verified trading system with audited live returns. (github.com, github.com) For readers scrolling past the post, the practical takeaway is narrower than the sales pitch: this is a public notebook that shows how quants combine market data, text analysis, and neural networks in one workflow. It is free to inspect, but the public record around it shows that reproducing the full setup may still require code, data, or assumptions that are not fully packaged in the repository. (x.com, github.com)