# -*- coding: utf-8 -*-
"""炉层杂质预警无监督模型冒烟脚本（Issue #71）。

直接运行 ``python _sanity_check_model.py`` 验证：ZScoreScorer 可在正常段 fit、
异常段产出高分数、ThresholdRule 触发预警、提前量评估为正（对齐 PRD 提前≥30min）。
零第三方依赖。
"""
import importlib.util
import os
import sys

_PKG_DIR = os.path.dirname(os.path.abspath(__file__))


def _load_pkg(name, path):
    if name in sys.modules:
        return
    spec = importlib.util.spec_from_file_location(
        name, os.path.join(path, "__init__.py"),
        submodule_search_locations=[path])
    mod = importlib.util.module_from_spec(spec)
    sys.modules[name] = mod
    spec.loader.exec_module(mod)


_load_pkg("impurity_forecast", _PKG_DIR)

from impurity_forecast import (  # noqa: E402
    FeatureVectorLike,
    ImpurityForecaster,
    ThresholdRule,
    ZScoreScorer,
)


def vec(ts, **kw):
    return FeatureVectorLike(timestamp=ts, values=dict(kw))


def main() -> int:
    # 正常段：炉温 850±5 波动，氯气 100±3 波动
    normal = [vec(i, 炉温=850.0 + (i % 3) * 2, 氯气=100.0 + (i % 2))
              for i in range(40)]
    # 观测段：前 40 正常，之后急升温 + 氯气突降
    obs = list(normal) + [
        vec(40 + i, 炉温=860.0 + 6.0 * i, 氯气=95.0 - i) for i in range(20)
    ]
    anomaly_ts = 59.0  # 末尾为异常峰值

    f = ImpurityForecaster(rule=ThresholdRule(
        score_threshold=3.0,
        feature_thresholds={"炉温": 900.0}))
    f.fit(normal)
    decisions, lt = f.evaluate(obs, anomaly_ts=anomaly_ts)

    triggered = [d for d in decisions if d.triggered]
    assert triggered, "异常段应触发预警"
    print(f"[OK] 预警触发 {len(triggered)} 次")
    print(f"[OK] 首次预警 ts={lt.first_alert_ts}，真实异常 ts={anomaly_ts}，"
          f"提前量={lt.lead_minutes:.1f} min（>0 即满足提前量口径）")

    # 模型可序列化
    d = f.scorer.to_dict()
    sc2 = ZScoreScorer.from_dict(d)
    s1 = f.scorer.score(obs[-1:])
    s2 = sc2.score(obs[-1:])
    assert abs(s1[0] - s2[0]) < 1e-9, "序列化前后分数应一致"
    print("[OK] 模型序列化往返一致（可版本化保存）")
    print("炉层杂质预警模型冒烟通过 ✅")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())