src/deploy_to_hf.py

"""
Helper script to push the Streamlit app and deployment files
to a Hugging Face Space.

This script mirrors the behaviour of the notebook's `hosting.py`:
- Ensures the Space exists (creating it if necessary),
- Uploads only the files required for deployment (code, Dockerfile,
  and requirements), excluding data, MLflow artifacts, etc.
"""

from __future__ import annotations

import os

from huggingface_hub import HfApi

import config


def main() -> None:
    token = config.HF_TOKEN or os.getenv("HF_TOKEN")
    if not token:
        raise ValueError(
            "HF_TOKEN is not set. Please export HF_TOKEN or configure it in config.py."
        )

    space_repo = config.HF_SPACE_REPO or os.getenv("HF_SPACE_REPO")
    if not space_repo:
        raise ValueError(
            "HF_SPACE_REPO is not set. Set it as an environment variable or in config.py."
        )

    api = HfApi(token=token)

    # Create the Space if it does not exist
    # Use "docker" SDK since we're deploying with Dockerfile
    api.create_repo(
        repo_id=space_repo,
        repo_type="space",
        space_sdk="docker",
        exist_ok=True,
    )

    # Create Space README with proper configuration
    space_readme = """---
title: Engine Predictive Maintenance
emoji: 🔧
colorFrom: blue
colorTo: red
sdk: docker
app_file: src/app.py
pinned: false
---

# Engine Predictive Maintenance System

Predictive maintenance application for engine failure classification using sensor data.

## Features

- Real-time engine condition prediction
- Interactive sensor visualizations
- Model inference from trained Random Forest classifier
- Modern Streamlit interface

## Usage

Enter sensor values (RPM, pressures, temperatures) to get real-time predictions about engine health.

## Model

Trained Random Forest model with hyperparameter tuning, deployed from Hugging Face Model Hub.
"""
    
    # Upload Space README first
    api.upload_file(
        path_or_fileobj=space_readme.encode(),
        path_in_repo="README.md",
        repo_id=space_repo,
        repo_type="space",
    )

    # Upload project files needed for deployment.
    # We ignore large local artifacts like raw data, mlruns, and models,
    # similar to how the reference notebook uploads only the deployment folder.
    api.upload_folder(
        folder_path=str(config.PROJECT_ROOT),
        path_in_repo=".",
        repo_id=space_repo,
        repo_type="space",
        ignore_patterns=[
            "data/*",
            "mlruns/*",
            "models/*",
            ".git/*",
            "__pycache__/*",
            ".github/*",
            "README.md",  # Don't upload project README, use Space-specific one
        ],
    )

    print(f"Deployment files pushed to Hugging Face Space: {space_repo}")


if __name__ == "__main__":
    main()