#!/usr/bin/env python3
"""Pick the winning configuration out of the model-selection run.

Ranking rule, fixed in advance so the choice cannot be tuned after the fact:
the mean CSI over the weeks the targets apply to (W4, W5, W6) in the
`daily_capable` framing, which is the strictest framing in which the stated
targets are not mathematically out of reach.  Ties break on the weekly-event
CSI, then on AUC.  The raw EC benchmark is ranked alongside but never selected.
"""

from __future__ import annotations

import json
import os
import sys

import numpy as np

sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
import ehd_common as C

TARGET_WEEKS = ["W4", "W5", "W6"]


def mean_metric(entry, frame, metric, weeks=TARGET_WEEKS):
    vals = [entry.get(frame, {}).get(w, {}).get(metric, np.nan) for w in weeks]
    vals = [v for v in vals if v is not None and np.isfinite(v)]
    return float(np.mean(vals)) if vals else -np.inf


def rank(results):
    scored = []
    for tag, entry in results.items():
        scored.append((
            mean_metric(entry, "daily_capable", "CSI"),
            mean_metric(entry, "weekly_all", "CSI"),
            mean_metric(entry, "daily_capable", "AUC"),
            tag))
    scored.sort(reverse=True)
    return scored


def main():
    path = (sys.argv[1] if len(sys.argv) > 1
            else os.path.join(C.OUT, "AI_Results", "model_selection.json"))
    with open(path) as fh:
        results = json.load(fh)
    scored = rank(results)

    sys.stderr.write("%-22s %9s %9s %9s\n"
                     % ("config", "CSI_cap", "CSI_week", "AUC_cap"))
    for csi, wcsi, auc, tag in scored:
        sys.stderr.write("%-22s %9.3f %9.3f %9.3f\n" % (tag, csi, wcsi, auc))

    for csi, wcsi, auc, tag in scored:
        if tag.startswith("ec_raw"):
            continue
        model, fset = tag.split("/")
        print("%s %s" % (model, fset))
        return
    print("lgbm_clf F3")


if __name__ == "__main__":
    main()