WeatherAI

Production-oriented Python application for weather feature prediction and classification using neural networks. The system learns from daily historical records (1997–2015), forecasts year-over-year feature trajectories, and classifies conditions as Thunderstorm, Rainy, Foggy, or Sunny.

Highlights

• Optimized ML pipeline — vectorized dataset construction, float32 features, scaled regression targets, Adam + early stopping.
• Honest metrics — temporal hold-out evaluation before the final fit (MAE, R², classification accuracy).
• Desktop studio UI — CustomTkinter dashboard with metrics, activity log, theme toggle, and cross-platform plot folder access.
• Tested — pytest suite with ≥90% coverage (pytest --cov).

Requirements

• Python 3.10+
• Dependencies in requirements.txt

Install

pip install -r requirements.txt

Run

On macOS with Homebrew Python, install Tk once (required for the GUI):

brew install python-tk@3.14   # match your python3 --version minor release

# Desktop app (recommended)
python desktop_app.py

# CLI engine
python weather_engine.py

Tests & coverage

pytest
bash scripts/run_coverage.sh

Coverage is enforced at ≥90% on the weather_ai package (engine, metrics, controller, visualizer). The CustomTkinter layout module (weather_ai/app.py) is excluded from the gate because headless CI often lacks _tkinter; UI flows are covered indirectly via AppController tests.

On Python 3.14, use scripts/run_coverage.sh (per-test coverage processes) instead of pytest --cov, which can conflict with NumPy’s import hooks.

Project layout

weather_ai/          # Core package (engine, metrics, UI, visualizer)
tests/               # Unit tests
WeatherYYYY.txt      # Training data (1997–2015)
desktop_app.py       # App entry point
weather_engine.py    # CLI entry point (backward compatible)

Dataset

Each WeatherYYYY.txt row contains a bias term, 11 numeric weather features, and a 4-character class code (0001, 0010, 0100, 1000).