Market Muse

An upgraded Streamlit dashboard for stock forecasting with a more thoughtful UI and a stronger modeling workflow than the original close-only LSTM demo.

What’s Better Now

• Multivariate LSTM training with technical signals like returns, moving averages, volatility, momentum, and RSI
• Baseline comparison against a naive previous-close model so forecast quality has context
• Confidence bands for forward forecasts
• Clearer diagnostics with RMSE, MAE, MAPE, and directional accuracy
• A redesigned interface with a guided overview, diagnostics tab, and forecast desk
• Exportable forecast table for quick downstream use

App Experience

The app is now organized into three parts:

1. Overview Read recent price action, liquidity, volatility, and the model’s forward path.
2. Diagnostics Compare the LSTM against a naive baseline and inspect training history plus residual drift.
3. Forecast Desk Review the forecast table, narrative insights, and download the result as CSV.

Run Locally

pip install -r requirements.txt
streamlit run main.py

Apple Silicon Note

If you are on an M-series Mac and see low-level TensorFlow crashes such as mutex.cc or mutex lock failed, recreate the environment with the Apple-specific packages from requirements.txt:

rm -rf .env
python3 -m venv .env
source .env/bin/activate
pip install --upgrade pip
pip install -r requirements.txt
streamlit run main.py

Controls

• Quick ticker: fast entry for popular symbols or a custom symbol
• Lookback window: how many past sessions the model sees in each sample
• Forecast horizon: how many future business days to project
• Training epochs: max epochs before early stopping steps in
• Train split: proportion of data used for model fitting before evaluation

Modeling Notes

• Historical data comes from Yahoo Finance via yfinance
• The model uses engineered technical features derived from price history
• Training uses early stopping and learning-rate reduction
• Forecast bands are heuristic ranges based on test-set residual dispersion

Important Caveat

This project is for experimentation and education. It is not financial advice, and the forecast ranges are not guarantees.