defect-analysis/defect_analysis/ml/features.py

"""结构化 ML 特征工程。"""

import pandas as pd

from defect_analysis.schemas import normalize_defect_schema


CATEGORICAL_FEATURES = [
    "equipment_id",
    "seat_id",
    "lam_equipment_id",
    "lam_seat_id",
    "lam_fixture_id",
    "lam_nozzle_id",
    "material_lot_oca",
    "material_lot_glass",
    "clean_equipment_id",
    "bond_equipment_id",
    "shift",
    "defect_geometry_type",
]

NUMERIC_FEATURES = [
    "x_mm",
    "y_mm",
    "hour",
    "width_mm",
    "height_mm",
    "length_mm",
    "angle_deg",
    "area_mm2",
]


def build_feature_frame(df):
    """把标准缺陷数据转换为可训练的数值特征矩阵。"""
    normalized = normalize_defect_schema(df)
    numeric = normalized[[col for col in NUMERIC_FEATURES if col in normalized.columns]].copy()
    for column in numeric.columns:
        numeric[column] = pd.to_numeric(numeric[column], errors="coerce").fillna(0.0)

    categorical = normalized[[col for col in CATEGORICAL_FEATURES if col in normalized.columns]].fillna("")
    encoded = pd.get_dummies(categorical.astype(str), prefix_sep="=", dtype=float)

    features = pd.concat([numeric.reset_index(drop=True), encoded.reset_index(drop=True)], axis=1)
    return features.fillna(0.0)