#!/usr/bin/env python3
"""step27_predictive_scan.py — 预测性指标扫描（真正的由果及因）
核心：不是事后分类，而是事前预测。
对全市场所有股票-日，每个12维因子，扫描各取值区间的"未来大涨概率"
预测目标：fwd_ret60 >= 50%（未来60日累计涨50% = 大涨段）
找事前有预测力的因子组合 → 形成信号
"""
import numpy as np
import pandas as pd

print("=== 加载 ===", flush=True)
panel = pd.read_pickle("/tmp/panel_12d.pkl")
panel = panel.sort_values(["code", "date"]).reset_index(drop=True)
panel["fwd_ret60"] = panel.groupby("code")["close"].transform(lambda x: x.shift(-60)/x - 1) * 100
panel["flow5_q"] = panel.groupby("date")["flow5"].transform(lambda x: x.rank(pct=True))
print("面板:", len(panel), flush=True)

# 预测目标：未来大涨
panel["is_big_future"] = (panel["fwd_ret60"] >= 50).astype(int)
base = panel.dropna(subset=["fwd_ret60"])
base_rate = base["is_big_future"].mean() * 100
base_avg = base["fwd_ret60"].mean()
print("全市场基线: 大涨率={:.2f}% avg60={:.2f}%".format(base_rate, base_avg), flush=True)
print("样本:", len(base), flush=True)

out = ["# 预测性指标扫描（事前预测，非事后分类）\n",
       "- 预测目标: 未来60日累计收益 >= 50%（大涨段）",
       "- 基线: 全市场大涨率 {:.2f}%\n".format(base_rate),
       "## 各因子分位扫描（大涨率 vs 基线）\n"]

def scan_factor(series, label, bins=None, is_rank=True):
    """扫描因子分位/取值区间的大涨率"""
    print("\n=== {} ===".format(label), flush=True)
    print("| 区间 | n | 大涨率 | avg60 | 超额 |", flush=True)
    print("|---|---:|---:|---:|---:|", flush=True)
    s = base[series].dropna()
    if len(s) < 1000:
        print("样本不足", flush=True)
        return
    if is_rank:
        # 分位区间
        qs = [0, 0.2, 0.4, 0.6, 0.8, 1.0]
        edges = s.quantile(qs).values
        for i in range(len(qs)-1):
            lo, hi = edges[i], edges[i+1]
            m = base[(base[series] >= lo) & (base[series] <= hi if i == len(qs)-2 else base[series] < hi)]
            if len(m) < 100:
                continue
            rate = m["is_big_future"].mean() * 100
            avg = m["fwd_ret60"].mean()
            out.append("| [{:.2f},{:.2f}) | {} | {:.2f}% | {:.2f}% | {:+.2f}pp |".format(
                lo, hi, len(m), rate, avg, rate - base_rate))
            print("| [{:.2f},{:.2f}) | {} | {:.2f}% | {:.2f}% | {:+.2f}pp |".format(
                lo, hi, len(m), rate, avg, rate - base_rate), flush=True)
    else:
        # 绝对值区间
        for lo, hi in bins:
            m = base[(base[series] >= lo) & (base[series] < hi)]
            if len(m) < 100:
                continue
            rate = m["is_big_future"].mean() * 100
            avg = m["fwd_ret60"].mean()
            out.append("| [{},{}) | {} | {:.2f}% | {:.2f}% | {:+.2f}pp |".format(
                lo, hi, len(m), rate, avg, rate - base_rate))
            print("| [{},{}) | {} | {:.2f}% | {:.2f}% | {:+.2f}pp |".format(
                lo, hi, len(m), rate, avg, rate - base_rate), flush=True)

# 技术面
scan_factor("rsi", "个股RSI")
scan_factor("bias60", "个股bias60")
scan_factor("bias20", "个股bias20")
scan_factor("dist_lo20", "距20日低点")
scan_factor("ret1", "当日涨幅")
scan_factor("ret5", "5日涨幅")
scan_factor("ret20", "20日涨幅")
scan_factor("vol_ratio", "量能比v5/v20")
# 大盘
scan_factor("mkt_rsi", "大盘RSI")
scan_factor("mkt_adx", "大盘ADX")
scan_factor("mkt_ret20", "大盘20日收益")
# 行业
scan_factor("sec_ret20", "行业20日动量")
# 资金
scan_factor("flow5_q", "资金流5日分位")
# 基本面
scan_factor("mcap_q", "市值分位")
scan_factor("pe_q", "PE分位")
scan_factor("pb_q", "PB分位")
# 消息
scan_factor("news3", "3日新闻数", is_rank=False, bins=[(0,1),(1,2),(2,4),(4,8),(8,100)])

with open("/tmp/step27_predictive_report.md", "w", encoding="utf-8") as f:
    f.write("\n".join(out))
print("\n报告已写 /tmp/step27_predictive_report.md", flush=True)
print("=== 完成 ===", flush=True)