tahamajs/ML/Files/ThisTerm/final_project - Storage Bucket

Traditional Persian Music — Classification & Clustering

This repository contains an end-to-end skeleton for the course project "Traditional Persian Music Classification and Clustering".

Structure

• data/ — place your raw .mp3 files here, organized as instrument_dastgah/filename.mp3 or similar.
• src/ — Python modules for preprocessing, feature extraction, modeling, and clustering.
• run_pipeline.py — example script to extract features and train baseline models.
• requirements.txt — Python dependencies.
• notebooks/ — (empty) place for EDA and report notebooks.

Quickstart

1. Create a virtual environment and install dependencies: python -m venv .venv && . .venv/bin/activate && pip install -r requirements.txt (If you need plotting / keras: pip install seaborn tensorflow)
2. Put your .mp3 files under data/ (one folder per Dastgah, see project spec).
3. Extract features and create features.csv: python run_pipeline.py --input_dir data --out_csv features.csv
4. Train baseline models (from extracted features): python run_pipeline.py --input_dir data --out_csv features.csv --train

Example: full pipeline with model saving

1. Run the end-to-end example (extract features, train models, clustering, save artifacts):

   python examples/full_pipeline.py --data_dir data --save_models_dir models
   

2. Perform inference on a single file using the helper:

   from src.inference import predict_file
   label, details = predict_file("data/shur/song1.mp3", "models")
   print(label, details)
   

MLflow (optional)

• To track experiments with MLflow, install mlflow and call tracking functions:

  from src.mlflow_tracking import log_classic_experiment, log_lstm_history
  log_classic_experiment("svm_experiment", params={"C":1.0}, metrics={"f1":0.78}, artifacts_dir="experiments")
  

• Start the MLflow UI locally: mlflow ui --port 5000

Notes

• Follow the course specification: each student collects 35 pieces (5 per Dastgah).
• See src/ for functions and examples to customize feature extraction and modeling.