#专题：估值分位计算方法

import numpy as np
import pandas as pd

def calc_percentile(current_value, history_series):
    """
    计算当前值在历史序列中的分位
    :param current_value: 当前 PE/PB 值
    :param history_series: 历史样本序列（pd.Series）
    :return: 分位百分比（0-100）
    """
    history = history_series.dropna().values
    if len(history) < 50:
        return np.nan  # 样本不足
    return np.percentile(history, current_value * 100 / current_value)  # 占位，见下方修正


def calc_percentile_correct(current_value, history_series):
    """
    正确实现：当前值在历史序列中的百分位
    """
    history = history_series.dropna().values
    if len(history) < 50:
        return np.nan
    # 方法1：用 sum 比例
    rank = (history <= current_value).sum() / len(history)
    return rank * 100


def calc_percentile_numpy(current_value, history_series):
    """
    用 numpy.percentile 的反向计算更简洁
    """
    history = history_series.dropna().values
    if len(history) < 50:
        return np.nan
    # scipy.stats.percentileofscore 的等效实现
    return 100.0 * np.sum(history <= current_value) / len(history)


def calc_index_pe_percentile(pe_series, window_years=10, freq='W'):
    """
    计算指数 PE 的历史分位（完整流程）
    :param pe_series: 日频 PE 序列，index 为日期
    :param window_years: 滚动窗口年数
    :param freq: 重采样周期 'W'=周 'M'=月
    :return: 当前分位
    """
    # 1. 重采样到周频（取每周收盘值）
    pe_resampled = pe_series.resample(freq).last()

    # 2. 取最近 window_years 年作为窗口
    cutoff_date = pe_resampled.index.max() - pd.DateOffset(years=window_years)
    window = pe_resampled[pe_resampled.index >= cutoff_date]

    # 3. 异常值处理：剔除负值与极端值
    window = window[(window > 0) & (window < 300)]

    # 4. 计算当前值的分位
    current_pe = pe_resampled.iloc[-1]
    percentile = 100.0 * np.sum(window.values <= current_pe) / len(window)

    return {
        'current_pe': current_pe,
        'percentile': percentile,
        'window_samples': len(window),
        'window_min': window.min(),
        'window_max': window.max(),
        'window_median': window.median()
    }


# 等权 PE 分位计算（多成分股）
def calc_equal_weight_pe_pe_percentile(stock_pe_df, date_col='date',
                                        code_col='code', pe_col='pe_ttm'):
    """
    计算等权 PE 的历史分位
    :param stock_pe_df: 包含 date/code/pe_ttm 的 DataFrame
    :return: 每个日期的等权 PE 序列
    """
    # 1. 按日期分组，计算每日等权 PE
    daily_eq_pe = (
        stock_pe_df.dropna(subset=[pe_col])
        .groupby(date_col)[pe_col]
        .mean()  # 等权 = 算术平均
    )

    # 2. 重采样到周频
    daily_eq_pe = daily_eq_pe.resample('W').last()

    # 3. 计算分位（同上）
    return calc_index_pe_percentile(daily_eq_pe)


# 使用示例
if __name__ == '__main__':
    # 模拟 10 年周频 PE 数据
    np.random.seed(42)
    dates = pd.date_range('2014-01-01', '2024-01-01', freq='W')
    pe_values = np.random.lognormal(mean=3.3, sigma=0.2, size=len(dates))
    pe_series = pd.Series(pe_values, index=dates)

    result = calc_index_pe_percentile(pe_series)
    print(f"当前 PE: {result['current_pe']:.2f}")
    print(f"历史分位: {result['percentile']:.1f}%")
    print(f"窗口样本数: {result['window_samples']}")
    print(f"区间: [{result['window_min']:.2f}, {result['window_max']:.2f}]")
    print(f"中位数: {result['window_median']:.2f}")