Chan/research/step4_zs_source_ab.py

"""Step 4：中枢来源 A/B —— cal_bi_zs(seg_list) vs cal_bi_zs_list_pure(bi_list)。

假设：seg 口径下笔中枢必须等所属线段成形，多叠了一层确认滞后；
pure 口径直接在扁平笔序列上滚动，应当显著更快出信号。
用同一套 walk-forward 重放，只切换中枢来源，对比滞后与真实收益。
"""
from __future__ import annotations

import argparse
import sys
from pathlib import Path

import numpy as np
import pandas as pd

sys.path.insert(0, str(Path(__file__).resolve().parent))

from lib.bsp_eval import baseline_stats, run_pipeline
from lib.data import fetch_ohlcv
from lib.walkforward import analyze_stability, replay

pd.set_option("display.width", 240)

from chanlun.core.ChanEnum import Chan_BSP_DIR

HORIZONS = (1, 3, 5, 10, 20, 40)


def evaluate(df: pd.DataFrame, tf: str, symbol: str, zs_source: str, window: int, step: int):
    _, final_bsp = run_pipeline(df, tf, zs_source=zs_source)
    final_keys = {
        (str(b.type).replace("Chan_BSP_TYPE.", ""),
         1 if b.dir == Chan_BSP_DIR.BUY else -1, str(b.end_time))
        for b in final_bsp if b.is_sure and b.sure_time is not None
    }
    result = replay(
        df, tf, window=window, step=step, zs_source=zs_source,
        cache_key=f"{symbol.replace('/', '_').replace(':', '-')}_{tf}",
    )
    life = analyze_stability(result, final_keys, df, window=window)
    valid = life[~life["truncated"]].dropna(subset=["fx_idx"]).copy()
    valid["fx_idx"] = valid["fx_idx"].astype(int)
    mature = valid[~valid["immature"]]
    return final_keys, valid, mature


def rt_returns(valid: pd.DataFrame, df: pd.DataFrame) -> pd.DataFrame:
    """以实时首见时刻入场的方向调整收益。"""
    closes = df["close"].to_numpy(dtype=float)
    n = len(df)
    rows = []
    for _, r in valid.iterrows():
        i, d = int(r["first_seen_idx"]), int(r["direction"])
        row = {"bsp_type": r["bsp_type"], "direction": d}
        for h in HORIZONS:
            j = i + h
            row[f"ret_{h}"] = d * (closes[j] - closes[i]) / closes[i] if j < n else np.nan
        rows.append(row)
    return pd.DataFrame(rows)


def stats_block(fwd: pd.DataFrame, df: pd.DataFrame, mask=None, label="") -> list[dict]:
    base = baseline_stats(df, HORIZONS).set_index("horizon")
    g = fwd if mask is None else fwd[mask]
    out = []
    for h in HORIZONS:
        r = g[f"ret_{h}"].dropna().to_numpy()
        if len(r) == 0:
            continue
        dirs = g.loc[g[f"ret_{h}"].notna(), "direction"].to_numpy()
        sd = r.std(ddof=1) if len(r) > 1 else np.nan
        out.append({
            "group": label, "horizon": h, "n": len(r), "mean": r.mean(),
            "winrate": (r > 0).mean(),
            "excess": r.mean() - float(np.mean(dirs) * base.loc[h, "base_mean_long"]),
            "tstat": r.mean() / (sd / np.sqrt(len(r))) if sd else np.nan,
        })
    return out


def fmt(rows: list[dict]) -> str:
    d = pd.DataFrame(rows)
    if d.empty:
        return "(无数据)"
    d["mean"] = d["mean"].map(lambda v: f"{v * 100:+.2f}%")
    d["excess"] = d["excess"].map(lambda v: f"{v * 100:+.2f}%")
    d["winrate"] = d["winrate"].map(lambda v: f"{v * 100:.0f}%")
    d["tstat"] = d["tstat"].map(lambda v: f"{v:+.2f}")
    return d.to_string(index=False)


def main() -> None:
    ap = argparse.ArgumentParser()
    ap.add_argument("--symbol", default="BTC/USDT:USDT")
    ap.add_argument("--tf", default="1h")
    ap.add_argument("--window", type=int, default=3000)
    ap.add_argument("--step", type=int, default=4)
    args = ap.parse_args()

    df = fetch_ohlcv(args.symbol, args.tf, 50000)
    print(f"[data] {args.symbol} {args.tf} rows={len(df)} "
          f"{df['date'].iloc[0]} -> {df['date'].iloc[-1]}\n")

    store = {}
    for src in ("seg", "pure"):
        print(f"===== 中枢来源: {src} =====")
        final_keys, valid, mature = evaluate(df, args.tf, args.symbol, src, args.window, args.step)
        fwd = rt_returns(valid, df)
        store[src] = (final_keys, valid, mature, fwd)
        print(f"  全量口径买卖点 = {len(final_keys)}")
        print(f"  重放有效信号   = {len(valid)}（已成熟 {len(mature)}）")
        print(f"  实时首见滞后   = 中位数 {valid['observed_lag'].median():.0f} 根 / "
              f"均值 {valid['observed_lag'].mean():.1f} 根")
        print(f"  幻影率         = {(~mature['in_final']).mean() * 100:.1f}%")
        print(f"  存活度         = {mature['persist_ratio'].mean():.3f}\n")

    print("\n########## 滞后对比 ##########")
    for src in ("seg", "pure"):
        v = store[src][1]
        q = v["observed_lag"].quantile([0.25, 0.5, 0.75])
        print(f"  {src:5s}  P25={q[0.25]:5.0f}  中位数={q[0.5]:5.0f}  P75={q[0.75]:5.0f}  n={len(v)}")

    print("\n########## 分类型滞后中位数 ##########")
    comp = pd.DataFrame({
        src: store[src][1].groupby("bsp_type")["observed_lag"].median()
        for src in ("seg", "pure")
    })
    comp["改善(根)"] = comp["seg"] - comp["pure"]
    print(comp.to_string())

    print("\n########## 分类型信号数量与幻影率 ##########")
    cnt = pd.DataFrame({
        f"{src}_n": store[src][1].groupby("bsp_type").size() for src in ("seg", "pure")
    })
    ph = pd.DataFrame({
        f"{src}_幻影": store[src][2].groupby("bsp_type")["in_final"].apply(lambda s: (1 - s.mean()) * 100)
        for src in ("seg", "pure")
    })
    print(pd.concat([cnt, ph], axis=1).round(1).to_string())

    print("\n########## 无未来函数收益：全部信号 ##########")
    for src in ("seg", "pure"):
        print(f"--- {src} ---")
        print(fmt(stats_block(store[src][3], df, None, "ALL")))

    print("\n########## 无未来函数收益：仅做多 ##########")
    for src in ("seg", "pure"):
        fwd = store[src][3]
        print(f"--- {src} ---")
        print(fmt(stats_block(fwd, df, fwd.direction == 1, "LONG")))

    print("\n########## 无未来函数收益：仅做空 ##########")
    for src in ("seg", "pure"):
        fwd = store[src][3]
        print(f"--- {src} ---")
        print(fmt(stats_block(fwd, df, fwd.direction == -1, "SHORT")))

    print("\n########## pure 口径分类型（持有 5 / 20 根）##########")
    fwd = store["pure"][3]
    rows = []
    for t in sorted(fwd["bsp_type"].unique()):
        rows += stats_block(fwd, df, fwd.bsp_type == t, t)
    d = pd.DataFrame(rows)
    d = d[d["horizon"].isin([5, 20])]
    print(fmt(d.to_dict("records")))


if __name__ == "__main__":
    main()