ProjectAeroMed — Exoplanet ML Research Tool (Improved)

Elevator pitch (2–3 lines): ProjectAeroMed is a reproducible ML pipeline and Streamlit app that analyzes transit data to classify candidate exoplanet signals using NASA datasets (Kepler/TESS). The project emphasizes reproducibility, explainability (SHAP), and deployability for quick follow-up prioritization.

---

What this repo contains (high level)

• scripts/get_nasa_data.py — script to download and preprocess NASA Exoplanet Archive / Kepler/TESS sample (usage shown below).
• train.py — training entrypoint with Stratified K-Fold cross-validation, model selection, and model metadata saving.
• exoplanet_model.py — model utilities: fit/predict/evaluate.
• app.py — Streamlit app with sidebar controls, model selection, metrics, and SHAP explainability placeholders.
• requirements.txt — pinned dependencies for reproducibility.
• Dockerfile, Makefile, Procfile — deployment helpers.
• models/ — recommended location for saved models (created by training).
• notebooks/ — optional EDA and results (add your own).

---

Data used

This project is designed to use NASA public datasets (examples):

• NASA Exoplanet Archive (Kepler Objects of Interest). Recommended CSV export URL (example): https://exoplanetarchive.ipac.caltech.edu/docs/API_kepcandidate_columns.html
• TESS candidate tables (via the Exoplanet Archive)

Important: The scripts/get_nasa_data.py script will attempt to download a CSV from a given URL and perform safe preprocessing. If you are offline, you can place a CSV at data/sample_kepler.csv (this repo includes a small sample). Always cite dataset URLs and the date you accessed them in your submission.

---

Quickstart (local)

1. Create a Python env and install dependencies:

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. (Optional) Download NASA sample data:

python scripts/get_nasa_data.py --url "https://your-nasa-csv-url.csv" --out data/nasa_kepler.csv

3. Train a model (example):

python train.py --data data/sample_kepler.csv --out models/best_model.joblib --cv 5

This will save models/best_model.joblib and models/model_metadata.json with metrics and hyperparameters. 4. Run the app:

streamlit run app.py

---

Reproducibility & notes

• Random seed is fixed in training to improve reproducibility.
• Model metadata (dataset checksum, hyperparameters, CV metrics mean/std) are saved alongside the model.
• For deployment, see Dockerfile and Procfile (Streamlit Cloud / Heroku).

---

Limitations & next steps

• Add automated unit tests for preprocessing and edge cases.
• Integrate live NASA API queries and scheduled retraining for new candidate ingestion.
• Add human-in-the-loop review and priority queuing for follow-up telescope observations.

---

Contact

Project by the ProjectAeroMed team — contact: projectaeromed@gmail.com Generated/updated: 2025-10-05T09:37:53.519973 UTC