# WashData - Home Assistant integration for appliance cycle monitoring via smart plugs.
# Copyright (C) 2026 Lukas Bandura
# SPDX-License-Identifier: AGPL-3.0-or-later
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as published
# by the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <https://www.gnu.org/licenses/>.
"""Auto-generated by ml_washdata/wash_ml/promotion.py. Do not edit by hand.

Embedded WashData model: 'hybrid_curve_quality' (target: 'problem_cycle').
Kind: standardized logistic regression. Runtime dependency: NumPy only.

Regenerate with ``./ml.sh experiment`` in the ml_washdata lab and copy the new
file. Determinism check at generation time: max_abs_score_diff=6.8331e-08
over 373 rows.

Usage in the integration::

    from .hybrid_curve_quality_model import score, predict, FEATURE_COLUMNS
    features = build_runtime_features(...)  # must populate FEATURE_COLUMNS
    is_positive = predict(features)
"""

from __future__ import annotations

import base64
import gzip
import json
from typing import Mapping

import numpy as np

MODEL_NAME = 'hybrid_curve_quality'
MODEL_TARGET = 'problem_cycle'
MODEL_KIND = 'standardized_logistic'
TARGET_UNITS = ''
# Training artifact for select_threshold only — NOT used at live inference.
# The live quality-suspicious gate reads ML_QUALITY_SUSPICIOUS_THRESHOLD from const.py.
THRESHOLD = 0.19

FEATURE_COLUMNS = [
    'duration_log_ratio',
    'energy_log_ratio',
    'peak_log_ratio',
    'profile_distance',
    'label_margin_positive',
    'max_gap_ratio',
    'low_power_gap_ratio',
    'false_end_energy_ratio',
    'sample_density_log',
    'peak_density_log',
    'local_spike_score',
    'local_spike_rate',
    'local_noise_score',
    'leading_idle_ratio',
    'trailing_idle_ratio',
    'trimmed_duration_log_ratio',
    'flag_pressure',
    'shape_fit_penalty',
    'shape_active_fraction',
    'shape_early_energy_fraction',
    'shape_late_energy_fraction',
    'shape_mid_trough_depth',
    'shape_peak_density',
    'shape_max_step_drop',
    'shape_max_step_rise',
    'shape_active_cv',
    'shape_autocorr_lag1',
    'shape_derivative_sign_changes',
    'shape_plateau_ratio',
    'shape_tail_slope',
    'has_trace',
]

# Provenance (metrics at training time):
MODEL_METRICS = json.loads("""{
    "owner_holdout": {
        "accuracy": 0.872,
        "balanced_accuracy": 0.89997,
        "f1": 0.898734,
        "fn": 15,
        "fp": 1,
        "positive_rate": 0.576,
        "precision": 0.986111,
        "problem_recall": 0.825581,
        "rows": 125,
        "specificity": 0.974359,
        "tn": 38,
        "tp": 71
    },
    "synthetic_all": {
        "accuracy": 1.0,
        "balanced_accuracy": 0.5,
        "f1": 1.0,
        "fn": 0,
        "fp": 0,
        "positive_rate": 1.0,
        "precision": 1.0,
        "problem_recall": 1.0,
        "rows": 1255,
        "specificity": 0.0,
        "tn": 0,
        "tp": 1255
    }
}""")

_MODEL_BLOB = (
    'H4sIAAAAAAACA3VW247bOAz9FcOvzaS6X7JPi6L7tBeg230qCkNjK4lR2/L60uls0X/fQzmTmem0yENkiiIPDymRX8vbNszlQe+V'
    '1txoo3dlHYclTuXhww3bM2GMsZbvaO0151Zua6mtMMLt2N5aJ6URBkshmJLcqh0pC6ad87SU0mqns5QLeIE5vudeSyNNPmaMwo+W'
    'UljpLfebrpQqGxPeCuW1yFKulJN5mSFYTXg4Z5o5pXZib5y2XAkyYTkThI5MKCaga7FUWnLvLWHnkjvpHCl4ox3XnqTMCuGV4CTV'
    'QltpCZqHKas84eFcGkjJhZMGMq/JLnBLZjJgphwXVmSqaO2yMY99ydVHcJziEQwTT84J5mmbawk4hkwxnJYe9NJ5w5UGYGKHG4Ms'
    '2Cx2nkGZkHEBsreAmBZcARwRLBzywTI9nkllPMvKiAO8kRfhPNg22YuwzIktSVwII9nmhHEH6s2WcoUAlATF0nBmrCLQSkoHpxek'
    'xiuWQ2UMgGymUCKLcG22oiACsmkkCV5UzgKVhMxpktbCpdrKzXkOg7SWYFttYu457NqNL2etyim1wiKTlzIFPKcFsTzFsMSmCkt5'
    'KAVq+YaZG+HfC3ZQ7qD4K8YOjJW78gi9dYpVnbq1H3AfPpTNOoWlTUPVpVOVl9CLQ5xO989EYwyfngumdGy7WDXtvIShjhB14TZ2'
    'VR+mUztUY5rbpf1M8j58qU5hvB7t0h227+L0THoM3RyrODTVxf3Dxhz6kRzFARYzqgc8z0VdqkNXzWP7KVZznab4nQz2HkVDaucn'
    'ajE07XCq2qaLV7/LFNruB9K270H2D4k7duFUjVOc5zXbnc9hjNWxXaoxDqFb7q+yUBM51XGiRRqu8him7v6BgRe7HUL46WbfNtUy'
    'pfV0BjHjcr5uPOXqURtJmZc4Vs2UxpfSCfR8D7b+/ChZlwTuJiA68au0iVP7OWTduT0NVX0OwynOj0AIf1gfM5ulC2iu5i6N5PAc'
    'ZgQRUE8o7BOIq1Pft1TYhungjIDOp3ZoIKC6a8LUtP8hHcgCCrGtqdwiclSjur+W6Q5cVefUNWldSBDqGnmr78sD3jQ8XOVt6Kh6'
    'cXme7uAFsMgm3z6clQpfQ3ng6BvHEf8owEt9b2V1oJtu6FbEup0pK5B4Zzjn+arcdrGvsBe6LtsUGs/wrpzSHXCiW4CLESePbU05'
    'orNWSe1RbrAkHf7h1fJv0LsflnNEpFW29TQkvmc/CUhvwWQFioNtYbAXYWSNJ0Fcvp/jz8Ir9JfY2QabbahJ5RuAD6GHg/J8fzuh'
    'UIENTv9dQ7cV5ZCWmB+kN6m/bYfYFLkKbnKJFK+Ky3Pzuk7o2l+W4oLppr6vu1g0cYn1kqY9LL1bh6XtY9GvM9TSuFLRFb+9/fX9'
    'P+/eVm/++v2fP/78uzhOqS86BF5kC68v9osmLKG4jUe8CwW9Drj/ZPQ93gIClYbiDrepoPq/ye/dXIzdOhfXtBSEcErd/EuxtKfz'
    'EnGiXc7Fee3DcDPFz228g6HL2YundY7TzTHU8EaR9Yku9p6uAOp2XHELLsNKJvcim5GNh5Rd1ui0Gi3G0uzB9gb/uW9YJZzFLIN2'
    'KRyXCk2StiVnaBa5J6NdcbHNI2jeArWXJxoMCyjVTdc6oahvOpzSioTU29CPbFbFWki3zTYavSkb0IxhmtC5+aEJW0vNDxJcKSGA'
    'RqFlWp/nK8EdBpNsjNoaV1vvRjBOcDqGhu8xKG1LWKCeTy4waUlBUsOkMyoPMRqtmsamTRfDyzZ5YBPjk89LULKNTByTj9xGG8ZB'
    'HQaAXZ4IMTF595G4Pcc+oHTvwnym+thvKcId61MTu9f0AC7oepGeqYfLksvqulGtQ7ugvkuSnNEg6E2idHL/7X9yriqGnQoAAA=='
)

_MODEL_CACHE: dict | None = None


def _load() -> dict:
    global _MODEL_CACHE
    if _MODEL_CACHE is None:
        payload = gzip.decompress(base64.b64decode(_MODEL_BLOB.encode("ascii")))
        spec = json.loads(payload.decode("utf-8"))
        _MODEL_CACHE = {
            "center": np.asarray(spec["center"], dtype=float),
            "scale": np.asarray(spec["scale"], dtype=float),
            "coef": np.asarray(spec["coef"], dtype=float),
            "bias": float(spec["bias"]),
            "threshold": float(spec["threshold"]),
            "output_center": float(spec.get("output_center") or 0.0),
            "output_scale": float(spec.get("output_scale") if spec.get("output_scale") is not None else 1.0),
            "feature_columns": list(spec["feature_columns"]),
        }
    return _MODEL_CACHE


def score(features: Mapping[str, float]) -> float:
    """Return the model probability in [0, 1] for one feature mapping."""
    model = _load()
    vector = np.array(
        [float(features.get(column) or 0.0) for column in model["feature_columns"]],
        dtype=float,
    )
    scaled = (vector - model["center"]) / model["scale"]
    logit = float(scaled @ model["coef"] + model["bias"])
    logit = max(-60.0, min(60.0, logit))
    return 1.0 / (1.0 + np.exp(-logit))


def predict(features: Mapping[str, float]) -> bool:
    """True when the example crosses the embedded decision threshold."""
    return score(features) >= _load()["threshold"]