供应链信号不确定性量化 — 从点预测到区间估计，支撑置信度分层决策

"""
供应链信号不确定性量化系统
功能：Conformal预测区间 / Bayesian深度集成 / 时序自适应UQ / Palantir置信度Tag
输入：时间序列预测模型 + 校准数据
输出：预测区间 + 置信度Tag + 决策分层建议
"""
import numpy as np
from dataclasses import dataclass
from typing import Optional
import warnings
warnings.filterwarnings('ignore')


@dataclass
class PredictionInterval:
    """预测区间——直接映射到Palantir Tag"""
    point_estimate: float
    lower_bound: float
    upper_bound: float
    coverage_level: float      # 目标覆盖率（如0.90）
    actual_coverage: float     # 实际覆盖率
    interval_width: float
    relative_uncertainty: float  # interval_width / point_estimate
    
    # Palantir Tag字段
    confidence_tier: str       # HIGH/MEDIUM/LOW
    recommended_action_tier: str  # AUTO/GUIDED/STAGED
    
    def to_palantir_tags(self) -> dict:
        return {
            "predicted_value": self.point_estimate,
            "prediction_interval_lower": self.lower_bound,
            "prediction_interval_upper": self.upper_bound,
            "prediction_confidence": self.coverage_level,
            "uncertainty_tier": self.confidence_tier,
            "action_tier": self.recommended_action_tier,
            "relative_uncertainty": round(self.relative_uncertainty, 4),
        }


class ConformalPredictionEngine:
    """
    供应链专用Conformal预测引擎
    保证：任何分布下的预测覆盖率 ≥ 1-α
    """
    
    def __init__(self, base_model, alpha: float = 0.10):
        """
        alpha: 显著性水平（0.10 = 90%覆盖率）
        """
        self.base_model = base_model
        self.alpha = alpha
        self.calibration_scores = []
        self._fitted = False
    
    def calibrate(self, X_cal: np.ndarray, y_cal: np.ndarray):
        """用校准数据计算非一致性得分"""
        predictions = self.base_model.predict(X_cal)
        # 非一致性得分：绝对误差
        self.calibration_scores = np.abs(y_cal - predictions)