#!/usr/bin/env python3
"""RipBench semantic-segmentation evaluator (framework-agnostic).

HOW TO USE (no CLI args - edit CONFIG below, then: python evaluate_semantic.py
in any env with numpy). Consumes the <stem>_confmat_<split>.npz files that
predict_semantic.py writes (per-image 4x4 GT-x-pred pixel confusion
matrices), sums them per split, and appends one TSV row per model x split to
eval/results/semantic_segmentation_eval.tsv.

No threshold protocol - semantic predictions are argmax maps, so every
metric is threshold-free: per-class IoU (rip_current first - the headline),
mIoU, pixel accuracy, mean per-class accuracy (macc). Per-video breakdowns
for the insights analysis aggregate the SAME npz files by video id.
"""
import glob
import os
_RIPBENCH_ROOT = os.environ.get("RIPBENCH_ROOT", os.path.normpath(os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "..")))  # release root (env RIPBENCH_ROOT overrides)
from datetime import datetime

import numpy as np

# --- CONFIG -------------------------------------------------------------------
PRED_DIR = _RIPBENCH_ROOT + "/models/semantic_segmentation/best_models/predictions"
SPLITS = ("val", "test")
OUT_TSV = os.path.join(os.path.dirname(os.path.abspath(__file__)), "results",
                       "semantic_segmentation_eval.tsv")
# ------------------------------------------------------------------------------

CLASSES = ("rip_current", "water", "land", "sky")  # train ids 0..3

HEADER = ("timestamp\tmodel\ttask\tsplit\tn_images"
          "\tmIoU\tIoU_rip_current\tIoU_water\tIoU_land\tIoU_sky"
          "\tpixel_acc\tmacc\n")


def metrics(conf):
    """conf: [4,4] float64, rows GT, cols pred."""
    tp = np.diag(conf)
    fp = conf.sum(0) - tp
    fn = conf.sum(1) - tp
    denom = tp + fp + fn
    iou = np.where(denom > 0, tp / np.maximum(denom, 1), np.nan)
    acc = np.where(conf.sum(1) > 0, tp / np.maximum(conf.sum(1), 1), np.nan)
    return {"iou": iou, "miou": float(np.nanmean(iou)),
            "pixel_acc": float(tp.sum() / conf.sum()),
            "macc": float(np.nanmean(acc))}


def main():
    os.makedirs(os.path.dirname(OUT_TSV), exist_ok=True)
    write_header = not os.path.isfile(OUT_TSV)
    pairs = {}
    for f in sorted(glob.glob(os.path.join(PRED_DIR, "*_confmat_*.npz"))):
        base = os.path.basename(f)
        stem, split = base.rsplit("_confmat_", 1)
        pairs.setdefault(stem, {})[split.removesuffix(".npz")] = f
    print(f"{len(pairs)} model(s) in {PRED_DIR}")

    with open(OUT_TSV, "a") as out:
        if write_header:
            out.write(HEADER)
        for stem, files in pairs.items():
            for split in SPLITS:
                if split not in files:
                    print(f"SKIP {stem}: no {split} confmat")
                    continue
                data = np.load(files[split], allow_pickle=False)
                conf = data["confmat"].sum(0).astype(np.float64)
                m = metrics(conf)
                row = [datetime.now().strftime("%Y-%m-%d %H:%M"), stem,
                       "semantic_segmentation", split, len(data["stems"]),
                       f"{m['miou']:.6f}"] + [f"{v:.6f}" for v in m["iou"]] + [
                       f"{m['pixel_acc']:.6f}", f"{m['macc']:.6f}"]
                out.write("\t".join(str(x) for x in row) + "\n")
                print(f"[{split}] {stem}: mIoU={m['miou']:.4f} "
                      f"rip-IoU={m['iou'][0]:.4f} pixAcc={m['pixel_acc']:.4f}")
    print(f"rows appended to {OUT_TSV}")


if __name__ == "__main__":
    main()