Audiolytics

An interactive Streamlit app that turns a single listening-history file into clear insights — genres over time, mood & energy, discovery habits, device mix, streaks, comebacks, and more — plus ML-powered 7-day forecasts.

What it does

• Timeline & habits: daily minutes, prime hours, streaks (calendar), seasonality.
• Genres: monthly share, variety (entropy), first/last appearance, rising vs. falling.
• Artists: comeback tracker (returns after X+ months), top artists.
• Mood & energy: valence × energy density map; rolling trends.
• Platform: regex-based platform normalization + Top-N + Other chart.
• Discovery & nostalgia: new vs repeat shares, “forgotten” past favorites.
• Extras: feature fingerprint & change between eras, optional session clustering.
• Forecasts (ML): 7-day platform-share forecasts with confidence bands and backtests.

How it works

• The app loads a prepared dataset and converts it to Parquet (if needed) for speed.
• Heavy transforms are cached; large views use sampling to stay responsive.
• Charts use Altair.
• ML forecasts are trained offline and read from Parquet at runtime by the Forecasts page.

ML at a glance

• Target: next-7-days platform share (normalized shares by day).
• Models: statsmodels SARIMAX / ETS baselines (per-platform series).
• Evaluation: rolling backtests (expanding origin), MAE/MAPE logged to disk.
• Artifacts: predictions with 80/95% intervals saved to Parquet for the app.

Deployment

• Works on Streamlit Cloud with no secrets at audiolytics.streamlit.app
• Main file: app/streamlit_app.py

Dataset (included)

This repo includes a dataset so the live app runs as-is:

• Primary: data/audiolytics.parquet (fastest to load)
• Optional: data/Complete_Data_with_genres_filled.csv (≈32 MB). It’s redundant for the app, but you can keep it for transparency. The app prefers the Parquet.

Source/enrichment: Kaggle base + cleaning + Spotify/Reccobeats audio features.

Dataset is included strictly so reviewers can experience the app end-to-end.

Getting started (local)

# clone
git clone https://github.com/gegedobruna/Audiolytics.git
cd Audiolytics

# (optional) create venv
python -m venv venv
# Windows: venv\Scripts\activate
# macOS/Linux: source venv/bin/activate

# install & run
pip install -r requirements.txt
streamlit run streamlit_app/app.py

---

License

Code: MIT.
Dataset: included for demo/review; please don’t redistribute outside this repo context.