Cuaca Prediction

Prediksi hujan (label biner Rain) dari fitur cuaca harian: temperatur, kelembapan, hujan (RR), penyinaran (SS), angin (FF_X, FF_AVG, DDD_X, DDD_CAR), serta tanggal (TANGGAL → Month/Day).

Diagram

Alur dari ingestion → cleaning → training → deployment:

Struktur Data

• Mentah (contoh iklim.csv / dataset.py): TANGGAL, TN, TX, TAVG, RH_AVG, RR, SS, FF_X, DDD_X, FF_AVG, DDD_CAR
• Saat training: cleaner menambah Month, Day, membuat label Rain = RR > 0, dan menjatuhkan RR.

Persiapan Lingkungan

Gunakan Python 3.12 (menghindari build pyarrow di 3.13):

python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

Atau:

make setup

Data

• Pakai data contoh: iklim.csv → split manual ke data/train.csv & data/test.csv via python dataset.py.
• Atau siapkan sendiri data/train.csv dan data/test.csv dengan skema mentah di atas.
• DVC (opsional): remote GDrive gdrive://1YbgrhKxIgpCDRrrMg72fK1Kwnj7q8cB4. Untuk service account:

  dvc remote modify --local myremote gdrive_use_service_account true
  dvc remote modify --local myremote gdrive_service_account_json_file_path .dvc/service-account.json
  dvc pull   # tarik data
  dvc push   # dorong cache

Train Model

python main.py          # atau make run

Model (default RandomForestClassifier) disimpan ke models/model.pkl. Logging & register MLflow ada di train_with_mlflow().

API (FastAPI)

uvicorn app:app --reload

Contoh payload POST /predict:

{
  "TANGGAL": "01-01-2025",
  "TN": 23.4,
  "TX": 29.2,
  "TAVG": 25.9,
  "RH_AVG": 92,
  "SS": 3.4,
  "FF_X": 5.0,
  "DDD_X": 330,
  "FF_AVG": 1.0,
  "DDD_CAR": "C"
}

Respons: {"predicted_class": 0|1} (0=tidak hujan, 1=hujan).

Docker

docker build -t weather-fastapi .
docker run -p 80:80 weather-fastapi

Monitoring

monitor.ipynb dan laporan drift (production_drift.html, test_drift.html) perlu disesuaikan ke skema cuaca; Evidently dapat diaktifkan via requirements (baris dikomentari).

Testing

python -m pytest   # atau make test

License

Copyright © 2024, Prasad Mahamulkar.

Released under the Apache-2.0 license.