#!/usr/bin/env python3
"""
2025年9月-2026年4月 每日新增注册人数回归分析（V4 - 比例潮汐+保底）
基于 daily_reg_analysis.py 的逻辑扩展时间范围
改进点：
1. 星期潮汐效应改为比例补偿
2. 实际值低于拟合值时保留实际值（活动日才用拟合值替换）
"""

import numpy as np
import pandas as pd
import sys

# ==================== 全量数据（2025-09-01 ~ 2026-04-30，来源：Online PostgreSQL） ====================
raw_data = {
    '2025-09-01': 3,   '2025-09-02': 10,  '2025-09-03': 4,   '2025-09-04': 5,
    '2025-09-05': 11,  '2025-09-06': 8,   '2025-09-07': 16,  '2025-09-08': 11,
    '2025-09-09': 137, '2025-09-10': 63,  '2025-09-11': 26,  '2025-09-12': 27,
    '2025-09-13': 41,  '2025-09-14': 39,  '2025-09-15': 27,  '2025-09-16': 57,
    '2025-09-17': 58,  '2025-09-18': 56,  '2025-09-19': 134, '2025-09-20': 104,
    '2025-09-21': 101, '2025-09-22': 135, '2025-09-23': 127, '2025-09-24': 71,
    '2025-09-25': 37,  '2025-09-26': 34,  '2025-09-27': 81,  '2025-09-28': 35,
    '2025-09-29': 47,  '2025-09-30': 30,
    '2025-10-01': 48,  '2025-10-02': 62,  '2025-10-03': 45,  '2025-10-04': 42,
    '2025-10-05': 45,  '2025-10-06': 62,  '2025-10-07': 42,  '2025-10-08': 46,
    '2025-10-09': 36,  '2025-10-10': 62,  '2025-10-11': 90,  '2025-10-12': 93,
    '2025-10-13': 162, '2025-10-14': 131, '2025-10-15': 112, '2025-10-16': 133,
    '2025-10-17': 215, '2025-10-18': 131, '2025-10-19': 81,  '2025-10-20': 45,
    '2025-10-21': 41,  '2025-10-22': 45,  '2025-10-23': 37,  '2025-10-24': 56,
    '2025-10-25': 80,  '2025-10-26': 50,  '2025-10-27': 89,  '2025-10-28': 87,
    '2025-10-29': 82,  '2025-10-30': 93,  '2025-10-31': 79,
    '2025-11-01': 129, '2025-11-02': 168, '2025-11-03': 77,  '2025-11-04': 68,
    '2025-11-05': 49,  '2025-11-06': 67,  '2025-11-07': 179, '2025-11-08': 161,
    '2025-11-09': 105, '2025-11-10': 219, '2025-11-11': 72,  '2025-11-12': 235,
    '2025-11-13': 104, '2025-11-14': 69,  '2025-11-15': 90,  '2025-11-16': 63,
    '2025-11-17': 70,  '2025-11-18': 82,  '2025-11-19': 156, '2025-11-20': 71,
    '2025-11-21': 93,  '2025-11-22': 71,  '2025-11-23': 92,  '2025-11-24': 47,
    '2025-11-25': 77,  '2025-11-26': 94,  '2025-11-27': 94,  '2025-11-28': 77,
    '2025-11-29': 123, '2025-11-30': 126,
    '2025-12-01': 91,  '2025-12-02': 95,  '2025-12-03': 191, '2025-12-04': 131,
    '2025-12-05': 125, '2025-12-06': 192, '2025-12-07': 195, '2025-12-08': 108,
    '2025-12-09': 118, '2025-12-10': 111, '2025-12-11': 104, '2025-12-12': 124,
    '2025-12-13': 190, '2025-12-14': 173, '2025-12-15': 98,  '2025-12-16': 101,
    '2025-12-17': 99,  '2025-12-18': 86,  '2025-12-19': 143, '2025-12-20': 127,
    '2025-12-21': 131, '2025-12-22': 69,  '2025-12-23': 77,  '2025-12-24': 109,
    '2025-12-25': 85,  '2025-12-26': 89,  '2025-12-27': 108, '2025-12-28': 96,
    '2025-12-29': 50,  '2025-12-30': 67,  '2025-12-31': 65,
    '2026-01-01': 78,  '2026-01-02': 74,  '2026-01-03': 69,  '2026-01-04': 43,
    '2026-01-05': 56,  '2026-01-06': 33,  '2026-01-07': 52,  '2026-01-08': 59,
    '2026-01-09': 58,  '2026-01-10': 84,  '2026-01-11': 75,  '2026-01-12': 35,
    '2026-01-13': 46,  '2026-01-14': 59,  '2026-01-15': 31,  '2026-01-16': 31,
    '2026-01-17': 67,  '2026-01-18': 71,  '2026-01-19': 53,  '2026-01-20': 48,
    '2026-01-21': 40,  '2026-01-22': 63,  '2026-01-23': 46,  '2026-01-24': 72,
    '2026-01-25': 86,  '2026-01-26': 61,  '2026-01-27': 57,  '2026-01-28': 149,
    '2026-01-29': 103, '2026-01-30': 87,  '2026-01-31': 61,
    '2026-02-01': 54,  '2026-02-02': 53,  '2026-02-03': 42,  '2026-02-04': 39,
    '2026-02-05': 42,  '2026-02-06': 55,  '2026-02-07': 36,  '2026-02-08': 47,
    '2026-02-09': 42,  '2026-02-10': 60,  '2026-02-11': 265, '2026-02-12': 59,
    '2026-02-13': 42,  '2026-02-14': 32,  '2026-02-15': 41,  '2026-02-16': 28,
    '2026-02-17': 52,  '2026-02-18': 23,  '2026-02-19': 35,  '2026-02-20': 26,
    '2026-02-21': 37,  '2026-02-22': 36,  '2026-02-23': 47,  '2026-02-24': 45,
    '2026-02-25': 64,  '2026-02-26': 180, '2026-02-27': 101, '2026-02-28': 168,
    '2026-03-01': 126, '2026-03-02': 106, '2026-03-03': 76,  '2026-03-04': 106,
    '2026-03-05': 451, '2026-03-06': 218, '2026-03-07': 241, '2026-03-08': 207,
    '2026-03-09': 168, '2026-03-10': 109, '2026-03-11': 133, '2026-03-12': 178,
    '2026-03-13': 244, '2026-03-14': 131, '2026-03-15': 132, '2026-03-16': 109,
    '2026-03-17': 95,  '2026-03-18': 96,  '2026-03-19': 77,  '2026-03-20': 103,
    '2026-03-21': 133, '2026-03-22': 126, '2026-03-23': 84,  '2026-03-24': 80,
    '2026-03-25': 94,  '2026-03-26': 85,  '2026-03-27': 95,  '2026-03-28': 105,
    '2026-03-29': 101, '2026-03-30': 81,  '2026-03-31': 99,
    '2026-04-01': 138, '2026-04-02': 164, '2026-04-03': 778, '2026-04-04': 341,
    '2026-04-05': 186, '2026-04-06': 207, '2026-04-07': 212, '2026-04-08': 751,
    '2026-04-09': 339, '2026-04-10': 128, '2026-04-11': 174, '2026-04-12': 151,
    '2026-04-13': 117, '2026-04-14': 133, '2026-04-15': 126, '2026-04-16': 115,
    '2026-04-17': 84,  '2026-04-18': 117, '2026-04-19': 120, '2026-04-20': 88,
    '2026-04-21': 99,  '2026-04-22': 179, '2026-04-23': 139, '2026-04-24': 122,
    '2026-04-25': 142, '2026-04-26': 137, '2026-04-27': 120, '2026-04-28': 163,
    '2026-04-29': 65,  '2026-04-30': 58,
}

dates = sorted(raw_data.keys())
date_series = pd.to_datetime(dates)
values = np.array([raw_data[d] for d in dates], dtype=float)
n = len(values)
x = np.arange(n, dtype=float)
dow = np.array([d.weekday() for d in date_series])  # 0=Mon, 6=Sun

# ==================== 活动标记 ====================
# 2026年1-4月（沿用原脚本的活动标记）
activity_events = {
    # 2025年9-12月活动（李承龙确认）
    '2025-09-09': ('活动', 0),
    '2025-09-10': ('活动', 0),
    '2025-09-19': ('活动', 0),
    '2025-09-20': ('活动', 0),
    '2025-09-21': ('活动', 0),
    '2025-09-22': ('活动', 0),
    '2025-09-23': ('活动', 0),
    '2025-10-13': ('活动', 0),
    '2025-10-14': ('活动', 0),
    '2025-10-16': ('活动', 0),
    '2025-10-17': ('活动', 0),
    '2025-11-02': ('活动', 0),
    '2025-11-07': ('活动', 0),
    '2025-11-10': ('活动', 0),
    '2025-11-12': ('活动', 0),
    '2025-11-19': ('活动', 0),
    '2025-12-03': ('活动', 0),
    # 2026年1-4月活动（沿用原标记）
    '2026-01-28': ('活动',    1),
    '2026-02-11': ('活动',    0),
    '2026-02-26': ('活动',    4),
    '2026-03-05': ('活动',    3),
    '2026-03-09': ('活动',    0),
    '2026-03-12': ('活动',    0),
    '2026-03-13': ('活动',    0),
    '2026-04-03': ('大型活动', 4),
    '2026-04-08': ('大型活动', 2),
    '2026-04-22': ('小活动',   1),
    '2026-04-28': ('小活动',   0),
}

affected_reason = {}
for d_str, (reason, aftereffect_days) in activity_events.items():
    base_idx = dates.index(d_str)
    affected_reason[d_str] = reason
    for offset in range(1, aftereffect_days + 1):
        idx = base_idx + offset
        if idx < n:
            ae_date = dates[idx]
            if ae_date not in affected_reason:
                affected_reason[ae_date] = f'{dates[base_idx]} 余波'

is_affected = np.array([d in affected_reason for d in dates])

# ==================== LOESS ====================
def loess_smooth(x_data, y_data, x_eval, frac=0.28, deg=2):
    n_eval = len(x_eval)
    n_data = len(x_data)
    y_smooth = np.zeros(n_eval)
    for j in range(n_eval):
        x0 = x_eval[j]; dist = np.abs(x_data - x0)
        k = max(int(frac * n_data), deg + 2); k = min(k, n_data)
        idx = np.argpartition(dist, k)[:k]
        x_nn, y_nn, d_nn = x_data[idx], y_data[idx], dist[idx]
        max_dist = np.max(d_nn)
        w = (1 - (d_nn / max_dist) ** 3) ** 3 if max_dist > 0 else np.ones(k)
        A = np.column_stack([np.ones(k), x_nn, x_nn**2]); W = np.diag(w)
        try:
            beta = np.linalg.solve(A.T @ W @ A, A.T @ W @ y_nn)
            y_smooth[j] = beta[0] + beta[1] * x0 + beta[2] * x0**2
        except np.linalg.LinAlgError:
            y_smooth[j] = np.average(y_nn, weights=w)
    return np.maximum(y_smooth, 0)

clean_mask = ~is_affected
clean_x, clean_y, clean_dow = x[clean_mask], values[clean_mask], dow[clean_mask]

# 第一轮：LOESS 拟合原始趋势
trend_raw = loess_smooth(clean_x, clean_y, x, frac=0.28, deg=2)

# ==================== 星期效应：比例方式 ====================
residual_pct = (clean_y - trend_raw[clean_mask]) / trend_raw[clean_mask]
dow_ratio = np.zeros(7)
for d in range(7):
    mask = clean_dow == d
    if np.sum(mask) > 0:
        dow_ratio[d] = np.mean(residual_pct[mask])

dow_ratio -= np.mean(dow_ratio)

print("=" * 80)
print("2025年9月-2026年4月 每日新增注册回归分析 V4（比例潮汐 + 保底）")
print("=" * 80)

print(f"\n📊 星期潮汐效应（比例方式）：")
dow_names = ['周一', '周二', '周三', '周四', '周五', '周六', '周日']
for i, name in enumerate(dow_names):
    bar = '█' * max(1, int(abs(dow_ratio[i]) * 200))
    sign = '+' if dow_ratio[i] >= 0 else ''
    print(f"  {name}: {sign}{dow_ratio[i]*100:.1f}% {bar}")

# ==================== 最终拟合 = 趋势 × (1 + 星期比例) ====================
trend_with_dow = trend_raw * (1 + dow_ratio[dow])
trend_with_dow = np.maximum(trend_with_dow, 0)

# ==================== 有效注册人数 ====================
effective = np.where(is_affected, np.minimum(values, trend_with_dow), values)

# ==================== 检测9-12月未标记的异常高峰 ====================
sep_start = dates.index('2025-09-01')
jan_start = dates.index('2026-01-01')

print(f"\n{'='*80}")
print("📋 完整结果（2025-09 ~ 2026-04）")
print(f"{'='*80}")
print(f"{'日期':>12} {'星期':>4} | {'实际':>5} | {'影响':>10} | {'趋势':>6} | {'趋势+潮汐':>9} | {'有效注册':>8} | {'剔除':>6}")
print("-" * 95)

for i in range(n):
    d = dates[i]
    dw = date_series[i].strftime('%a')
    tag = affected_reason.get(d, '—')
    trend_val = trend_raw[i]
    tdow_val = trend_with_dow[i]
    eff_val = effective[i]
    removed = values[i] - eff_val
    print(f"{d:>12} {dw:>4} | {values[i]:>5.0f} | {tag:>10} | {trend_val:>6.0f} | {tdow_val:>9.0f} | {eff_val:>8.0f} | {removed:>+6.0f}")

# ==================== 月度汇总 ====================
print(f"\n{'='*80}")
print("📊 月度汇总")
print(f"{'='*80}")
print(f"{'月份':>10} | {'实际总计':>10} | {'有效总计':>10} | {'剔除':>8} | {'剔除率':>7} | {'实际日均':>8} | {'有效日均':>8} | {'活动天数':>8}")
print("-" * 105)

months = [
    ('2025-09', '2025-09-01', '2025-10-01'),
    ('2025-10', '2025-10-01', '2025-11-01'),
    ('2025-11', '2025-11-01', '2025-12-01'),
    ('2025-12', '2025-12-01', '2026-01-01'),
    ('2026-01', '2026-01-01', '2026-02-01'),
    ('2026-02', '2026-02-01', '2026-03-01'),
    ('2026-03', '2026-03-01', '2026-04-01'),
    ('2026-04', '2026-04-01', '2026-05-01'),
]

all_monthly = []
for label, start, end in months:
    si = dates.index(start)
    ei = dates.index(end) if end in dates else n
    act = np.sum(values[si:ei])
    eff = np.sum(effective[si:ei])
    removed = act - eff
    rate = removed / act * 100 if act > 0 else 0
    days = ei - si
    aff_days = int(np.sum(is_affected[si:ei]))
    print(f"{label:>10} | {act:>10.0f} | {eff:>10.0f} | {removed:>8.0f} | {rate:>6.1f}% | {act/days:>8.1f} | {eff/days:>8.1f} | {aff_days:>8}")
    all_monthly.append({
        '月份': label, '实际总计': int(act), '有效总计': int(eff),
        '剔除': int(removed), '剔除率%': round(rate, 1),
        '实际日均': round(act/days, 1), '有效日均': round(eff/days, 1),
        '活动天数': aff_days
    })

# ==================== 整体汇总 ====================
total_actual = np.sum(values)
total_eff = np.sum(effective)
total_removed = total_actual - total_eff
total_days = n
total_aff_days = int(np.sum(is_affected))

print(f"\n{'='*80}")
print("📊 全期汇总（2025-09-01 ~ 2026-04-30）")
print(f"{'='*80}")
print(f"  统计天数:       {total_days}")
print(f"  活动影响天数:   {total_aff_days}")
print(f"")
print(f"  实际注册总计:   {total_actual:.0f}")
print(f"  有效注册总计:   {total_eff:.0f}")
print(f"  剔除活动量:     {total_removed:.0f} 人 ({total_removed/total_actual*100:.1f}%)")
print(f"")
print(f"  实际日均:       {np.mean(values):.1f}")
print(f"  有效日均:       {np.mean(effective):.1f}")

# 分段统计
sep_jan_act = np.sum(values[sep_start:jan_start])
sep_jan_eff = np.sum(effective[sep_start:jan_start])
jan_apr_act = np.sum(values[jan_start:])
jan_apr_eff = np.sum(effective[jan_start:])

print(f"")
print(f"  2025-09~12 实际: {sep_jan_act:.0f}  有效: {sep_jan_eff:.0f}  剔除: {sep_jan_act-sep_jan_eff:.0f} ({(sep_jan_act-sep_jan_eff)/sep_jan_act*100:.1f}%)")
print(f"  2026-01~04 实际: {jan_apr_act:.0f}  有效: {jan_apr_eff:.0f}  剔除: {jan_apr_act-jan_apr_eff:.0f} ({(jan_apr_act-jan_apr_eff)/jan_apr_act*100:.1f}%)")

# ==================== 9-12月未标记的异常高峰检测 ====================
print(f"\n{'='*80}")
print("⚠️ 2025年9-12月 可能的活动高峰（实际值 > 趋势+潮汐 × 1.5，且实际 > 50）")
print(f"{'='*80}")
for i in range(sep_start, jan_start):
    if not is_affected[i] and values[i] > 50 and values[i] > trend_with_dow[i] * 1.5:
        ratio = values[i] / trend_with_dow[i]
        print(f"  {dates[i]} ({date_series[i].strftime('%a')}): 实际={values[i]:.0f}, 拟合={trend_with_dow[i]:.0f}, 倍数={ratio:.1f}x")

# ==================== 导出Excel ====================
rows = []
for i in range(n):
    rows.append({
        '日期': dates[i],
        '星期': date_series[i].strftime('%A'),
        '实际注册人数': int(values[i]),
        '是否活动影响': '是' if is_affected[i] else '否',
        '活动说明': affected_reason.get(dates[i], ''),
        'LOESS趋势': round(trend_raw[i], 0),
        '趋势+潮汐(比例)': round(trend_with_dow[i], 0),
        '有效注册人数(用于转化)': round(effective[i], 0),
        '剔除人数': round(values[i] - effective[i], 0),
    })

output_df = pd.DataFrame(rows)
output_path = '/root/.openclaw/workspace/output/2025-09_2026-04_注册回归分析.xlsx'

with pd.ExcelWriter(output_path, engine='openpyxl') as writer:
    output_df.to_excel(writer, sheet_name='每日明细', index=False)
    
    # 月度汇总 sheet
    pd.DataFrame(all_monthly).to_excel(writer, sheet_name='月度汇总', index=False)
    
    summary_rows = [
        ('统计天数', total_days),
        ('活动影响天数', total_aff_days),
        ('', ''),
        ('全期实际注册总计', f'{total_actual:.0f}'),
        ('全期有效注册总计', f'{total_eff:.0f}'),
        ('全期剔除活动量', f'{total_removed:.0f}'),
        ('全期剔除比例', f'{total_removed/total_actual*100:.1f}%'),
        ('', ''),
        ('全期实际日均', f'{np.mean(values):.1f}'),
        ('全期有效日均', f'{np.mean(effective):.1f}'),
        ('', ''),
        ('2025-09~12 实际总计', f'{sep_jan_act:.0f}'),
        ('2025-09~12 有效总计', f'{sep_jan_eff:.0f}'),
        ('2025-09~12 剔除', f'{sep_jan_act-sep_jan_eff:.0f}'),
        ('', ''),
        ('2026-01~04 实际总计', f'{jan_apr_act:.0f}'),
        ('2026-01~04 有效总计', f'{jan_apr_eff:.0f}'),
        ('2026-01~04 剔除', f'{jan_apr_act-jan_apr_eff:.0f}'),
        ('', ''),
        ('星期效应-周一', f'{dow_ratio[0]*100:+.1f}%'),
        ('星期效应-周二', f'{dow_ratio[1]*100:+.1f}%'),
        ('星期效应-周三', f'{dow_ratio[2]*100:+.1f}%'),
        ('星期效应-周四', f'{dow_ratio[3]*100:+.1f}%'),
        ('星期效应-周五', f'{dow_ratio[4]*100:+.1f}%'),
        ('星期效应-周六', f'{dow_ratio[5]*100:+.1f}%'),
        ('星期效应-周日', f'{dow_ratio[6]*100:+.1f}%'),
    ]
    pd.DataFrame(summary_rows, columns=['指标','值']).to_excel(writer, sheet_name='汇总', index=False)

print(f"\n✅ Excel 已导出: {output_path}")