"""Integration and submission-contract tests for final_pipeline.py."""

from __future__ import annotations

import unittest
from pathlib import Path

import pandas as pd

import numpy as np

from benchmark_grouped import FAULT_LABELS, FREQ_COLS, NOT_APPLICABLE
from final_pipeline import (
    OOD_OK_TO_UNKNOWN_THRESHOLD_MV,
    OOD_OK_TO_UNKNOWN_RATIO_THRESHOLD,
    apply_ood_override,
    run_pipeline,
    validate_inference_data,
    validate_submission,
)


REPO_DIR = Path(__file__).resolve().parent


class FinalPipelineIntegrationTests(unittest.TestCase):
    @classmethod
    def setUpClass(cls) -> None:
        cls.val = pd.read_csv(REPO_DIR / "val.csv").reset_index(drop=True)
        cls.test = pd.read_csv(REPO_DIR / "test.csv").reset_index(drop=True)
        cls.sample = pd.read_csv(REPO_DIR / "sample_submit.csv").reset_index(drop=True)
        cls.submission, cls.diagnostics = run_pipeline(
            val=cls.val,
            test=cls.test,
            sample_submit=cls.sample,
            random_state=42,
            n_jobs=1,
        )

    def test_submission_contract_is_valid(self) -> None:
        validate_submission(self.submission, self.test)
        self.assertEqual(len(self.submission), 600)

    def test_severity_business_rule_is_enforced(self) -> None:
        fault = self.submission["label"].isin(FAULT_LABELS)
        self.assertTrue(
            self.submission.loc[~fault, "severity"].eq(NOT_APPLICABLE).all()
        )
        self.assertFalse(
            self.submission.loc[fault, "severity"].eq(NOT_APPLICABLE).any()
        )

    def test_diagnostics_are_complete_for_the_application(self) -> None:
        self.assertEqual(len(self.diagnostics), len(self.test))
        classifier = self.diagnostics["decision_source"].eq("classifier")
        ood = self.diagnostics["decision_source"].eq("ood_override")
        self.assertTrue(
            self.diagnostics.loc[classifier, "label_confidence"].between(0, 1).all()
        )
        self.assertTrue(self.diagnostics.loc[ood, "label_confidence"].isna().all())
        self.assertTrue(self.diagnostics["label_margin"].between(0, 1).all())
        self.assertTrue(
            self.diagnostics["anomaly_score_mean_abs_mv"].ge(0).all()
        )
        self.assertEqual(
            int(self.diagnostics["missing_spectral_cells"].sum()),
            int(self.test.filter(regex=r"^mV_").isna().sum().sum()),
        )

    def test_validator_rejects_wrong_non_fault_severity(self) -> None:
        broken = self.submission.copy()
        position = broken.index[~broken["label"].isin(FAULT_LABELS)][0]
        broken.loc[position, "severity"] = "male"
        with self.assertRaisesRegex(ValueError, "nie_dotyczy"):
            validate_submission(broken, self.test)


class InferenceInputTests(unittest.TestCase):
    def test_incomplete_engine_is_rejected(self) -> None:
        test = pd.read_csv(REPO_DIR / "test.csv").reset_index(drop=True)
        incomplete = test.drop(index=test.index[0]).reset_index(drop=True)
        with self.assertRaisesRegex(ValueError, "Incomplete"):
            validate_inference_data(incomplete)


class OODRuleTests(unittest.TestCase):
    def test_only_anomalous_ok_is_overridden(self) -> None:
        labels = np.asarray(["ok", "ok", "ok", "pompa"], dtype=object)
        features = np.zeros((4, 2 * len(FREQ_COLS)), dtype=float)
        features[2, len(FREQ_COLS) : 2 * len(FREQ_COLS)] = (
            OOD_OK_TO_UNKNOWN_THRESHOLD_MV + 1.0
        )
        features[3, len(FREQ_COLS) : 2 * len(FREQ_COLS)] = (
            OOD_OK_TO_UNKNOWN_THRESHOLD_MV + 10.0
        )
        predicted, overridden, _, ratio = apply_ood_override(
            labels,
            features,
            np.asarray(["engine_a", "engine_a", "engine_a", "engine_b"]),
        )
        self.assertGreater(ratio[2], OOD_OK_TO_UNKNOWN_RATIO_THRESHOLD)
        self.assertEqual(predicted.tolist(), ["ok", "ok", "unknown", "pompa"])
        self.assertEqual(overridden.tolist(), [False, False, True, False])

    def test_globally_noisy_engine_is_not_treated_as_isolated_ood(self) -> None:
        labels = np.asarray(["ok", "ok", "ok"], dtype=object)
        features = np.zeros((3, 2 * len(FREQ_COLS)), dtype=float)
        features[:, len(FREQ_COLS) : 2 * len(FREQ_COLS)] = 10.0
        predicted, overridden, score, ratio = apply_ood_override(
            labels,
            features,
            np.asarray(["noisy", "noisy", "noisy"]),
        )
        self.assertTrue((score > OOD_OK_TO_UNKNOWN_THRESHOLD_MV).all())
        self.assertTrue((ratio == 1.0).all())
        self.assertEqual(predicted.tolist(), labels.tolist())
        self.assertFalse(overridden.any())


if __name__ == "__main__":
    unittest.main(verbosity=2)