Phase C dry-run·문서화·DB 증분 저장 및 운영 env 동기화

- 1분봉 다운로드 제외, MONITOR_PERSIST로 05/06 수집 시 coins.db INSERT - Phase C paper_fires 로그·07 모의 리포트, hybrid 시뮬 산출물·reference 문서 갱신 - .env Phase C(LIVE=0), bootstrap dotenv override=True Co-authored-by: Cursor <cursoragent@cursor.com>
2026-06-01 23:32:47 +09:00
parent 3cbfa40aab
commit b9ee241d14
19 changed files with 877 additions and 333 deletions
--- a/scripts/07_phase_c_paper_report.py
+++ b/scripts/07_phase_c_paper_report.py
@@ -0,0 +1,165 @@
+#!/usr/bin/env python3
+"""
+Phase C dry-run 종료 후 모의 수익률(참고) 집계.
+
+- 입력: data/ops/paper_fires.jsonl (06 dry-run 발화 로그)
+- 출력: docs/05_ops/phase_c_paper_report.json + 콘솔 요약
+
+주의: 실계좌 수익이 아님. 발화가 N봉 후 가격으로 계산한 forward % 합산(참고).
+      hybrid 복리 PnL은 simulation_report.html 과 다릅니다.
+"""
+from __future__ import annotations
+
+import json
+import runpy
+from datetime import datetime
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+runpy.run_path(str(Path(__file__).resolve().parent / "_bootstrap.py"))
+
+from config import (  # noqa: E402
+    MATCH_FORWARD_BARS,
+    MATCH_PRIMARY_INTERVAL,
+    SYMBOL,
+    TRADING_FEE_RATE,
+)
+from deepcoin.matching.label_outcomes import _forward_ret_vectorized  # noqa: E402
+from deepcoin.ops.monitor import Monitor  # noqa: E402
+from deepcoin.paths import PAPER_FIRES_LOG, PAPER_WEEKLY_REPORT_JSON  # noqa: E402
+
+_FEE_PCT = TRADING_FEE_RATE * 2 * 100
+
+
+def load_paper_fires(path: Path) -> pd.DataFrame:
+    """paper_fires.jsonl → DataFrame."""
+    if not path.is_file():
+        return pd.DataFrame()
+    rows = []
+    for line in path.read_text(encoding="utf-8").splitlines():
+        line = line.strip()
+        if line:
+            rows.append(json.loads(line))
+    if not rows:
+        return pd.DataFrame()
+    return pd.DataFrame(rows)
+
+
+def attach_forward_returns(fires: pd.DataFrame, close_df: pd.DataFrame) -> pd.DataFrame:
+    """
+    would_trade=True 발화에 MATCH_FORWARD_BARS 기준 forward 수익률(%) 부여.
+
+    Args:
+        fires: paper 발화.
+        close_df: 3분 종가 (datetime index).
+
+    Returns:
+        forward_ret_pct 컬럼 추가.
+    """
+    if fires.empty or close_df.empty:
+        fires["forward_ret_pct"] = np.nan
+        return fires
+
+    close_df = close_df.sort_index()
+    if not isinstance(close_df.index, pd.DatetimeIndex):
+        close_df.index = pd.to_datetime(close_df.index)
+    close_ts_ns = close_df.index.astype(np.int64).values
+    close_px = close_df["Close"].astype(float).values
+
+    sub = fires[fires["would_trade"] == True].copy()  # noqa: E712
+    if sub.empty:
+        fires["forward_ret_pct"] = np.nan
+        return fires
+
+    sig = pd.to_datetime(sub["signal_dt"])
+    fire_ns = sig.astype(np.int64).values
+    c0 = sub["close"].astype(float).values
+    side = sub["side"].astype(str).values
+    ret, valid = _forward_ret_vectorized(
+        fire_ns, c0, close_ts_ns, close_px, side, MATCH_FORWARD_BARS, _FEE_PCT
+    )
+    fires = fires.copy()
+    fires["forward_ret_pct"] = np.nan
+    idx = sub.index
+    fires.loc[idx, "forward_ret_pct"] = np.where(valid, ret, np.nan)
+    return fires
+
+
+def summarize(fires: pd.DataFrame) -> dict:
+    """집계 dict."""
+    traded = fires[fires["would_trade"] == True]  # noqa: E712
+    with_ret = traded[traded["forward_ret_pct"].notna()]
+    out: dict = {
+        "generated_at": datetime.now().isoformat(timespec="seconds"),
+        "symbol": SYMBOL,
+        "forward_bars": MATCH_FORWARD_BARS,
+        "fee_round_trip_pct": _FEE_PCT,
+        "total_signals": int(len(fires)),
+        "would_trade_count": int(len(traded)),
+        "skipped_count": int(len(fires) - len(traded)),
+        "labeled_count": int(len(with_ret)),
+        "note": (
+            "모의 forward 수익률. 실계좌·hybrid 복리 PnL 아님. "
+            "매수·매도 leg 미결합 단순 합산."
+        ),
+    }
+    if not with_ret.empty:
+        out["mean_forward_ret_pct"] = round(float(with_ret["forward_ret_pct"].mean()), 4)
+        out["sum_forward_ret_pct"] = round(float(with_ret["forward_ret_pct"].sum()), 4)
+        by_side = (
+            with_ret.groupby("side")["forward_ret_pct"]
+            .agg(["count", "mean", "sum"])
+            .round(4)
+        )
+        out["by_side"] = {k: v.to_dict() for k, v in by_side.iterrows()}
+    by_rule = (
+        traded.groupby("rule_id")
+        .size()
+        .to_dict()
+        if not traded.empty
+        else {}
+    )
+    out["fires_by_rule"] = by_rule
+    return out
+
+
+def main() -> None:
+    """paper_fires 로드 → forward % → 리포트 저장."""
+    fires = load_paper_fires(PAPER_FIRES_LOG)
+    if fires.empty:
+        print(f"[07] 발화 로그 없음: {PAPER_FIRES_LOG}")
+        print("  Phase C 기간 06_execute_live.py (LIVE=0) 상시 실행 후 재시도")
+        return
+
+    mon = Monitor(cooldown_file=None)
+    df = mon.read_candles_from_db(SYMBOL, MATCH_PRIMARY_INTERVAL, max_rows=50000)
+    if df.empty:
+        df = mon.get_coin_some_data(SYMBOL, MATCH_PRIMARY_INTERVAL)
+    if not isinstance(df.index, pd.DatetimeIndex):
+        df = df.set_index(pd.to_datetime(df["datetime"]))
+
+    fires = attach_forward_returns(fires, df)
+    report = summarize(fires)
+    PAPER_WEEKLY_REPORT_JSON.parent.mkdir(parents=True, exist_ok=True)
+    PAPER_WEEKLY_REPORT_JSON.write_text(
+        json.dumps(report, ensure_ascii=False, indent=2),
+        encoding="utf-8",
+    )
+
+    print(f"[07] 저장: {PAPER_WEEKLY_REPORT_JSON}")
+    print(f"  기간 로그: {fires['ts'].min()} ~ {fires['ts'].max()}")
+    print(f"  발화 {report['total_signals']} · 체결가정(would_trade) {report['would_trade_count']}")
+    if "sum_forward_ret_pct" in report:
+        print(
+            f"  모의 forward 합산: {report['sum_forward_ret_pct']}% "
+            f"(평균 {report['mean_forward_ret_pct']}%, "
+            f"{MATCH_FORWARD_BARS}봉 후, 참고용)"
+        )
+    else:
+        print("  forward 라벨 가능 건 없음 (봉 데이터 부족 또는 발화 없음)")
+
+
+if __name__ == "__main__":
+    main()