gihyeon/cointrader
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(ml): pass SL/TP multipliers to dataset generation — align train/serve

Browse Source 
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
21in7
2026-03-21 18:16:50 +09:00
parent 75d1af7fcc
commit 0cc5835b3a
4 changed files with 41 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
scripts/train_mlx_model.py
View File

@@ -45,7 +45,7 @@ def _split_combined(df: pd.DataFrame) -> tuple[pd.DataFrame, pd.DataFrame | None
return xrp_df, btc_df, eth_df
def train_mlx(data_path: str, time_weight_decay: float = 2.0) -> float:
def train_mlx(data_path: str, time_weight_decay: float = 2.0, atr_sl_mult: float = 2.0, atr_tp_mult: float = 2.0) -> float:
print(f"데이터 로드: {data_path}")
raw = pd.read_parquet(data_path)
print(f"캔들 수: {len(raw)}")
@@ -58,7 +58,8 @@ def train_mlx(data_path: str, time_weight_decay: float = 2.0) -> float:
print("\n데이터셋 생성 중...")
t0 = time.perf_counter()
dataset = generate_dataset_vectorized(df, btc_df=btc_df, eth_df=eth_df, time_weight_decay=time_weight_decay)
dataset = generate_dataset_vectorized(df, btc_df=btc_df, eth_df=eth_df, time_weight_decay=time_weight_decay,
atr_sl_mult=atr_sl_mult, atr_tp_mult=atr_tp_mult)
t1 = time.perf_counter()
print(f"데이터셋 생성 완료: {t1 - t0:.1f}초, {len(dataset)}개 샘플")
@@ -170,6 +171,8 @@ def walk_forward_auc(
time_weight_decay: float = 2.0,
n_splits: int = 5,
train_ratio: float = 0.6,
atr_sl_mult: float = 2.0,
atr_tp_mult: float = 2.0,
) -> None:
"""Walk-Forward 검증: 슬라이딩 윈도우로 n_splits번 학습/검증 반복."""
print(f"\n=== Walk-Forward 검증 ({n_splits}폴드, decay={time_weight_decay}) ===")
@@ -177,7 +180,8 @@ def walk_forward_auc(
df, btc_df, eth_df = _split_combined(raw)
dataset = generate_dataset_vectorized(
df, btc_df=btc_df, eth_df=eth_df, time_weight_decay=time_weight_decay
df, btc_df=btc_df, eth_df=eth_df, time_weight_decay=time_weight_decay,
atr_sl_mult=atr_sl_mult, atr_tp_mult=atr_tp_mult,
)
missing = [c for c in FEATURE_COLS if c not in dataset.columns]
for col in missing:
@@ -260,12 +264,16 @@ def main():
)
parser.add_argument("--wf", action="store_true", help="Walk-Forward 검증 실행")
parser.add_argument("--wf-splits", type=int, default=5, help="Walk-Forward 폴드 수")
parser.add_argument("--sl-mult", type=float, default=2.0, help="SL ATR 배수 (기본 2.0)")
parser.add_argument("--tp-mult", type=float, default=2.0, help="TP ATR 배수 (기본 2.0)")
args = parser.parse_args()
if args.wf:
walk_forward_auc(args.data, time_weight_decay=args.decay, n_splits=args.wf_splits)
walk_forward_auc(args.data, time_weight_decay=args.decay, n_splits=args.wf_splits,
atr_sl_mult=args.sl_mult, atr_tp_mult=args.tp_mult)
else:
train_mlx(args.data, time_weight_decay=args.decay)
train_mlx(args.data, time_weight_decay=args.decay,
atr_sl_mult=args.sl_mult, atr_tp_mult=args.tp_mult)
if __name__ == "__main__":

26
scripts/train_model.py
View File

@@ -54,8 +54,6 @@ def _cgroup_cpu_count() -> int:
LOOKAHEAD = 24 # 15분봉 × 24 = 6시간 (dataset_builder.py와 동기화)
ATR_SL_MULT = 1.5
ATR_TP_MULT = 3.0
MODEL_PATH = Path("models/lgbm_filter.pkl")
PREV_MODEL_PATH = Path("models/lgbm_filter_prev.pkl")
LOG_PATH = Path("models/training_log.json")
@@ -63,6 +61,8 @@ LOG_PATH = Path("models/training_log.json")
def _process_index(args: tuple) -> dict | None:
"""단일 인덱스에 대해 피처+레이블을 계산한다. Pool worker 함수."""
ATR_SL_MULT = 1.5 # legacy values
ATR_TP_MULT = 3.0
i, df_values, df_columns = args
df = pd.DataFrame(df_values, columns=df_columns)
@@ -191,7 +191,7 @@ def _load_lgbm_params(tuned_params_path: str | None) -> tuple[dict, float]:
return lgbm_params, weight_scale
def train(data_path: str, time_weight_decay: float = 2.0, tuned_params_path: str | None = None):
def train(data_path: str, time_weight_decay: float = 2.0, tuned_params_path: str | None = None, atr_sl_mult: float = 2.0, atr_tp_mult: float = 2.0):
print(f"데이터 로드: {data_path}")
df_raw = pd.read_parquet(data_path)
print(f"캔들 수: {len(df_raw)}, 컬럼: {list(df_raw.columns)}")
@@ -218,6 +218,8 @@ def train(data_path: str, time_weight_decay: float = 2.0, tuned_params_path: str
df, btc_df=btc_df, eth_df=eth_df,
time_weight_decay=time_weight_decay,
negative_ratio=5,
atr_sl_mult=atr_sl_mult,
atr_tp_mult=atr_tp_mult,
)
if dataset.empty or "label" not in dataset.columns:
@@ -335,6 +337,8 @@ def walk_forward_auc(
n_splits: int = 5,
train_ratio: float = 0.6,
tuned_params_path: str | None = None,
atr_sl_mult: float = 2.0,
atr_tp_mult: float = 2.0,
) -> None:
"""Walk-Forward 검증: 슬라이딩 윈도우로 n_splits번 학습/검증 반복.
@@ -359,6 +363,8 @@ def walk_forward_auc(
df, btc_df=btc_df, eth_df=eth_df,
time_weight_decay=time_weight_decay,
negative_ratio=5,
atr_sl_mult=atr_sl_mult,
atr_tp_mult=atr_tp_mult,
)
actual_feature_cols = [c for c in FEATURE_COLS if c in dataset.columns]
X = dataset[actual_feature_cols].values
@@ -422,7 +428,7 @@ def walk_forward_auc(
print(f" 폴드별: {[round(a, 4) for a in aucs]}")
def compare(data_path: str, time_weight_decay: float = 2.0, tuned_params_path: str | None = None):
def compare(data_path: str, time_weight_decay: float = 2.0, tuned_params_path: str | None = None, atr_sl_mult: float = 2.0, atr_tp_mult: float = 2.0):
"""기존 피처 vs OI 파생 피처 추가 버전 A/B 비교."""
import warnings
@@ -449,6 +455,8 @@ def compare(data_path: str, time_weight_decay: float = 2.0, tuned_params_path: s
df, btc_df=btc_df, eth_df=eth_df,
time_weight_decay=time_weight_decay,
negative_ratio=5,
atr_sl_mult=atr_sl_mult,
atr_tp_mult=atr_tp_mult,
)
if dataset.empty:
@@ -546,6 +554,8 @@ def main():
)
parser.add_argument("--compare", action="store_true",
help="OI 파생 피처 추가 전후 A/B 성능 비교")
parser.add_argument("--sl-mult", type=float, default=2.0, help="SL ATR 배수 (기본 2.0)")
parser.add_argument("--tp-mult", type=float, default=2.0, help="TP ATR 배수 (기본 2.0)")
args = parser.parse_args()
# --symbol 모드: 심볼별 디렉토리 경로 자동 결정
@@ -563,16 +573,20 @@ def main():
args.data = "data/combined_15m.parquet"
if args.compare:
compare(args.data, time_weight_decay=args.decay, tuned_params_path=args.tuned_params)
compare(args.data, time_weight_decay=args.decay, tuned_params_path=args.tuned_params,
atr_sl_mult=args.sl_mult, atr_tp_mult=args.tp_mult)
elif args.wf:
walk_forward_auc(
args.data,
time_weight_decay=args.decay,
n_splits=args.wf_splits,
tuned_params_path=args.tuned_params,
atr_sl_mult=args.sl_mult,
atr_tp_mult=args.tp_mult,
)
else:
train(args.data, time_weight_decay=args.decay, tuned_params_path=args.tuned_params)
train(args.data, time_weight_decay=args.decay, tuned_params_path=args.tuned_params,
atr_sl_mult=args.sl_mult, atr_tp_mult=args.tp_mult)
if __name__ == "__main__":

9
scripts/tune_hyperparams.py
View File

@@ -39,7 +39,7 @@ from src.dataset_builder import generate_dataset_vectorized, stratified_undersam
# 데이터 로드 및 데이터셋 생성 (1회 캐싱)
# ──────────────────────────────────────────────
def load_dataset(data_path: str) -> tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]:
def load_dataset(data_path: str, atr_sl_mult: float = 2.0, atr_tp_mult: float = 2.0) -> tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]:
"""
parquet 로드 → 벡터화 데이터셋 생성 → (X, y, w) numpy 배열 반환.
study 시작 전 1회만 호출하여 모든 trial이 공유한다.
@@ -64,7 +64,8 @@ def load_dataset(data_path: str) -> tuple[np.ndarray, np.ndarray, np.ndarray, np
df = df_raw[base_cols].copy()
print("\n데이터셋 생성 중 (1회만 실행)...")
dataset = generate_dataset_vectorized(df, btc_df=btc_df, eth_df=eth_df, time_weight_decay=0.0, negative_ratio=5)
dataset = generate_dataset_vectorized(df, btc_df=btc_df, eth_df=eth_df, time_weight_decay=0.0, negative_ratio=5,
atr_sl_mult=atr_sl_mult, atr_tp_mult=atr_tp_mult)
if dataset.empty or "label" not in dataset.columns:
raise ValueError("데이터셋 생성 실패: 샘플 0개")
@@ -527,6 +528,8 @@ def main():
parser.add_argument("--train-ratio", type=float, default=0.6, help="학습 구간 비율 (기본: 0.6)")
parser.add_argument("--min-recall", type=float, default=0.35, help="최소 재현율 제약 (기본: 0.35)")
parser.add_argument("--no-baseline", action="store_true", help="베이스라인 측정 건너뜀")
parser.add_argument("--sl-mult", type=float, default=2.0, help="SL ATR 배수 (기본 2.0)")
parser.add_argument("--tp-mult", type=float, default=2.0, help="TP ATR 배수 (기본 2.0)")
args = parser.parse_args()
# --symbol 모드: 심볼별 디렉토리 경로 자동 결정
@@ -538,7 +541,7 @@ def main():
args.data = "data/combined_15m.parquet"
# 1. 데이터셋 로드 (1회)
X, y, w, source = load_dataset(args.data)
X, y, w, source = load_dataset(args.data, atr_sl_mult=args.sl_mult, atr_tp_mult=args.tp_mult)
# 2. 베이스라인 측정
if args.symbol:

2
src/backtester.py
View File

@@ -743,6 +743,8 @@ class WalkForwardBacktester:
signal_threshold=self.cfg.signal_threshold,
adx_threshold=self.cfg.adx_threshold,
volume_multiplier=self.cfg.volume_multiplier,
atr_sl_mult=self.cfg.atr_sl_mult,
atr_tp_mult=self.cfg.atr_tp_mult,
)
except Exception as e:
logger.warning(f" [{symbol}] 데이터셋 생성 실패: {e}")