"""End-to-end smoke test of the post-R0 pipeline.

Calls preprocess_image_and_predict on each bundled sample face. After
R0 the function uses the MediaPipe Face Landmarker (tasks API) for
face detection + bbox, and the upstream ResNet50 + Grad-CAM for
emotion classification. This exercises both halves of the new
pipeline together.

Run from repo root: `cd ~/Code/research/totes-emosh && uv run python scratch/smoke_upstream.py`
"""

from __future__ import annotations

import os
import sys
from pathlib import Path

from PIL import Image

ROOT = Path(__file__).parent.parent
os.chdir(ROOT)
sys.path.insert(0, str(ROOT))


def main() -> None:
    from app.app_utils import preprocess_image_and_predict

    images_dir = ROOT / "images"
    samples = sorted(
        p for p in images_dir.glob("*.png")
        if p.stem not in {"LMLLOGO", "LMLOBS"}
    )

    print(f"{'sample':<14} {'top-1':<14} {'p':>5}  {'top-2':<14} {'p':>5}  {'top-3':<14} {'p':>5}")
    print("-" * 90)

    for sample in samples:
        img = Image.open(sample).convert("RGB")
        face, _heatmap, confidences = preprocess_image_and_predict(img)
        if confidences is None:
            print(f"{sample.stem:<14} NO FACE")
            continue
        ranked = sorted(confidences.items(), key=lambda x: -x[1])
        top3 = ranked[:3]
        row = [sample.stem.ljust(14)]
        for label, p in top3:
            row.append(f"{label:<14} {p:.3f}")
        print("  ".join(row))


if __name__ == "__main__":
    main()