[2026-05-28] 외부 데이터 학습 파이프라인 복구

.gitignore

@@ -4,6 +4,7 @@ data/daily_context.json
 data/universe_cache.json
 data/training_dataset*.csv
 data/external_training_dataset*.csv
+data/external/
 data/kis_token_*.json
 data/news/
 data/market/

app/ml/features.py

@@ -15,16 +15,24 @@ EXCLUDED_COLUMNS = {
     "name",
     "entry_time",
     "exit_time",
+    "exit_price",
     "sample_time",
     "created_at",
     "exit_reason",
     "strategy",
     "reason",
+    "pnl",
     "source_file",
     "ai_win_score",
     "ai_stop_loss_score",
     "ai_model_version",
 }
+EXCLUDED_PREFIXES = (
+    "price_",
+    "ret_",
+    "mfe_",
+    "mae_",
+)
 
 
 def select_feature_columns(df: pd.DataFrame, targets: Iterable[str] = LABEL_COLUMNS) -> list[str]:
@@ -32,7 +40,11 @@ def select_feature_columns(df: pd.DataFrame, targets: Iterable[str] = LABEL_COLU
     numeric_columns = [
         column
         for column in df.columns
-        if column not in excluded and pd.api.types.is_numeric_dtype(df[column])
+        if (
+            column not in excluded
+            and not column.startswith(EXCLUDED_PREFIXES)
+            and pd.api.types.is_numeric_dtype(df[column])
+        )
     ]
     return sorted(numeric_columns)
 

scripts/build_external_training_dataset.py

@@ -31,16 +31,29 @@ def _num(value, default=0.0):
         return default
 
 
-def _load_daily_amounts() -> dict[tuple[str, str], dict]:
+def _load_daily_amounts() -> dict[str, list[dict]]:
-    result = {}
+    by_ticker = defaultdict(dict)
     for file in DAILY_ROOT.glob("*/stocks.csv"):
         rows = _read_csv(file)
         for row in rows:
             date = str(row.get("date") or file.parent.name)
             ticker = str(row.get("ticker") or row.get("티커") or "")
             if ticker:
-                result[(date, ticker)] = row
+                row["date"] = date
-    return result
+                by_ticker[ticker][date] = row
+    return {
+        ticker: [rows[date] for date in sorted(rows)]
+        for ticker, rows in by_ticker.items()
+    }
+
+
+def _previous_daily_row(daily: dict[str, list[dict]], date: str, ticker: str) -> dict:
+    previous = {}
+    for row in daily.get(ticker, []):
+        if str(row.get("date", "")) >= date:
+            break
+        previous = row
+    return previous
 
 
 def _future_metrics(rows: list[dict], idx: int, entry_price: float):
@@ -84,7 +97,7 @@ def _rows_for_file(path: Path, daily: dict, k: float, breakout_only: bool):
     for date in sorted(by_date):
         day_rows = by_date[date]
         ticker = day_rows[0]["ticker"]
-        daily_row = daily.get((date, ticker), {})
+        daily_row = _previous_daily_row(daily, date, ticker)
         prev_high = _num(daily_row.get("high"))
         prev_low = _num(daily_row.get("low"))
         prev_amount = _num(daily_row.get("amount"))

scripts/collect_daily_features.py

@@ -6,15 +6,23 @@ Outputs:
     data/external/daily/YYYYMMDD/indexes.csv
 """
 import argparse
+import asyncio
+import csv
+import os
 import sys
-from datetime import datetime
+from datetime import datetime, timedelta
 from pathlib import Path
 
 import pandas as pd
 
 
 ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT))
 OUT_ROOT = ROOT / "data" / "external" / "daily"
+ETF_KEYWORDS = (
+    "인버스", "레버리지", "선물", "KODEX", "TIGER", "KBSTAR",
+    "HANARO", "ARIRANG", "KOSEF", "SOL", "ACE", "RISE", "PLUS",
+)
 
 
 def _yyyymmdd(date_text: str | None) -> str:
@@ -44,6 +52,20 @@ def _standardize_index_row(row: dict, date_yyyymmdd: str, code: str, name: str)
     return out
 
 
+def _is_etf(ticker: str, name: str) -> bool:
+    if ticker.startswith("Q") or len(ticker) != 6:
+        return True
+    return any(keyword in name for keyword in ETF_KEYWORDS)
+
+
+def _write_csv(path: Path, rows: list[dict], fieldnames: list[str]):
+    path.parent.mkdir(parents=True, exist_ok=True)
+    with path.open("w", encoding="utf-8-sig", newline="") as f:
+        writer = csv.DictWriter(f, fieldnames=fieldnames)
+        writer.writeheader()
+        writer.writerows(rows)
+
+
 def collect_with_pykrx(date_yyyymmdd: str, out_dir: Path):
     try:
         from pykrx import stock
@@ -67,16 +89,73 @@ def collect_with_pykrx(date_yyyymmdd: str, out_dir: Path):
     return len(stocks), len(index_rows)
 
 
+async def collect_with_kis(date_yyyymmdd: str, out_dir: Path, top: int):
+    from app.main import load_env
+    from app.execution.kis_client import KISClient
+
+    load_env()
+    os.environ["KIS_MOCK"] = "false"
+
+    kis = KISClient()
+    await kis.get_access_token()
+
+    rank = await kis.get_volume_rank(top_n=top * 2)
+    tickers = [
+        r["ticker"] for r in rank
+        if not _is_etf(r["ticker"], r["name"])
+    ][:top]
+
+    end_dt = datetime.strptime(date_yyyymmdd, "%Y%m%d")
+    start = (end_dt - timedelta(days=14)).strftime("%Y%m%d")
+
+    stock_rows = []
+    for ticker in tickers:
+        try:
+            rows = await kis.get_ohlcv_daily(ticker, start=start, end=date_yyyymmdd)
+        except Exception as exc:
+            print(f"daily fetch failed {ticker}: {exc}", file=sys.stderr)
+            continue
+        for row in rows:
+            stock_rows.append({
+                "date": row["date"],
+                "ticker": ticker,
+                "open": row["open"],
+                "high": row["high"],
+                "low": row["low"],
+                "close": row["close"],
+                "volume": row["volume"],
+                "amount": row["close"] * row["volume"],
+            })
+        await asyncio.sleep(0.25)
+
+    _write_csv(
+        out_dir / "stocks.csv",
+        stock_rows,
+        ["date", "ticker", "open", "high", "low", "close", "volume", "amount"],
+    )
+    _write_csv(
+        out_dir / "indexes.csv",
+        [],
+        ["date", "code", "name", "open", "high", "low", "close", "volume", "amount", "change_pct"],
+    )
+    return len(stock_rows), 0
+
+
 def main():
     parser = argparse.ArgumentParser()
     parser.add_argument("--date", help="YYYY-MM-DD or YYYYMMDD. Defaults to today.")
+    parser.add_argument("--top", type=int, default=30)
     args = parser.parse_args()
 
     date_yyyymmdd = _yyyymmdd(args.date)
     out_dir = OUT_ROOT / date_yyyymmdd
     out_dir.mkdir(parents=True, exist_ok=True)
 
-    stock_count, index_count = collect_with_pykrx(date_yyyymmdd, out_dir)
+    try:
+        stock_count, index_count = collect_with_pykrx(date_yyyymmdd, out_dir)
+    except Exception as exc:
+        print(f"pykrx daily features failed, falling back to KIS: {exc}", file=sys.stderr)
+        stock_count, index_count = asyncio.run(collect_with_kis(date_yyyymmdd, out_dir, args.top))
     print(f"saved daily features: stocks={stock_count}, indexes={index_count}, dir={out_dir}")
 
 

scripts/collect_minute_data.py

@@ -23,6 +23,14 @@ from app.execution.kis_client import KISClient
 
 
 OUT_ROOT = ROOT / "data" / "external" / "minute"
+DEFAULT_HOURS = (
+    "093000", "100000", "103000", "110000", "113000",
+    "120000", "123000", "130000", "133000", "140000",
+)
+ETF_KEYWORDS = (
+    "인버스", "레버리지", "선물", "KODEX", "TIGER", "KBSTAR",
+    "HANARO", "ARIRANG", "KOSEF", "SOL", "ACE", "RISE", "PLUS",
+)
 
 
 def _date_dir(date_text: str | None) -> str:
@@ -39,6 +47,12 @@ def _load_cached_tickers(limit: int) -> list[str]:
     return list(data.get("tickers", []))[:limit]
 
 
+def _is_etf(ticker: str, name: str) -> bool:
+    if ticker.startswith("Q") or len(ticker) != 6:
+        return True
+    return any(keyword in name for keyword in ETF_KEYWORDS)
+
+
 async def _resolve_tickers(kis: KISClient, args) -> list[str]:
     if args.tickers:
         return [t.strip() for t in args.tickers.split(",") if t.strip()]
@@ -47,8 +61,12 @@ async def _resolve_tickers(kis: KISClient, args) -> list[str]:
     if cached:
         return cached
 
-    rank = await kis.get_volume_rank(top_n=args.top)
+    rank = await kis.get_volume_rank(top_n=args.top * 2)
-    return [r["ticker"] for r in rank]
+    tickers = [
+        r["ticker"] for r in rank
+        if args.include_etf or not _is_etf(r["ticker"], r["name"])
+    ]
+    return tickers[:args.top]
 
 
 def _write_csv(path: Path, rows: list[dict]):
@@ -59,6 +77,17 @@ def _write_csv(path: Path, rows: list[dict]):
         writer.writerows(rows)
 
 
+async def _collect_ticker_rows(kis: KISClient, ticker: str, hours: list[str], sleep: float) -> list[dict]:
+    by_key = {}
+    for hour in hours:
+        rows = await kis.get_ohlcv_minute(ticker, hour=hour)
+        for row in rows:
+            key = (row.get("date"), row.get("time"), row.get("ticker"))
+            by_key[key] = row
+        await asyncio.sleep(sleep)
+    return sorted(by_key.values(), key=lambda r: (r.get("date", ""), r.get("time", "")))
+
+
 async def main_async(args):
     load_env()
     if args.real_quotes:
@@ -67,12 +96,15 @@ async def main_async(args):
     kis = KISClient()
     await kis.get_access_token()
     tickers = await _resolve_tickers(kis, args)
+    hours = [h.strip() for h in (args.hours or args.hour or "").split(",") if h.strip()]
+    if not hours:
+        hours = list(DEFAULT_HOURS)
 
     out_dir = OUT_ROOT / _date_dir(args.date)
     saved = 0
     for ticker in tickers:
         try:
-            rows = await kis.get_ohlcv_minute(ticker, hour=args.hour)
+            rows = await _collect_ticker_rows(kis, ticker, hours, args.sleep)
             if rows:
                 _write_csv(out_dir / f"{ticker}.csv", rows)
                 saved += 1
@@ -81,7 +113,6 @@ async def main_async(args):
                 print(f"no rows {ticker}")
         except Exception as exc:
             print(f"failed {ticker}: {exc}", file=sys.stderr)
-        await asyncio.sleep(args.sleep)
 
     print(f"minute collection done: saved={saved}/{len(tickers)}, dir={out_dir}")
 
@@ -91,9 +122,11 @@ def main():
     parser.add_argument("--date", help="YYYY-MM-DD or YYYYMMDD. Defaults to today.")
     parser.add_argument("--tickers", help="Comma-separated tickers.")
     parser.add_argument("--top", type=int, default=30)
-    parser.add_argument("--hour", default="153000", help="KIS upper-bound time HHMMSS.")
+    parser.add_argument("--hour", default="", help="Single KIS upper-bound time HHMMSS.")
+    parser.add_argument("--hours", default=",".join(DEFAULT_HOURS), help="Comma-separated KIS upper-bound times.")
     parser.add_argument("--sleep", type=float, default=1.1)
     parser.add_argument("--real-quotes", action="store_true", help="Use real quote API even if .env is mock.")
+    parser.add_argument("--include-etf", action="store_true", help="Include ETF/ETN products in collection.")
     args = parser.parse_args()
     asyncio.run(main_async(args))
 

scripts/run_training_pipeline.ps1

@@ -49,7 +49,7 @@ Invoke-PythonStep -Name "collecting KIS minute data" -Args @("scripts\collect_mi
 
 Invoke-PythonStep -Name "exporting bot training dataset" -Args @("scripts\export_training_dataset.py", "data\training_dataset.csv") -Required $true
 
-Invoke-PythonStep -Name "building external training dataset" -Args @("scripts\build_external_training_dataset.py", "--out", "data\external_training_dataset.csv") -Required $true
+Invoke-PythonStep -Name "building external training dataset" -Args @("scripts\build_external_training_dataset.py", "--out", "data\external_training_dataset.csv", "--all-minutes") -Required $true
 
 Invoke-PythonStep -Name "training model" -Args @("scripts\train_ai_model.py") -Required $true