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

feat: enhance model training and deployment scripts with time-weighted sampling

Browse Source 
- Updated `train_model.py` and `train_mlx_model.py` to include a time weight decay parameter for improved sample weighting during training.
- Modified dataset generation to incorporate sample weights based on time decay, enhancing model performance.
- Adjusted deployment scripts to support new backend options and improved error handling for model file transfers.
- Added new entries to the training log for better tracking of model performance metrics over time.
- Included ONNX model export functionality in the MLX filter for compatibility with Linux servers.
This commit is contained in:
21in7
2026-03-01 21:25:06 +09:00
parent 301457ce57
commit db144750a3
10 changed files with 324 additions and 97 deletions
Download Patch File Download Diff File Expand all files Collapse all files

84
scripts/deploy_model.sh
View File

@@ -1,65 +1,77 @@
#!/usr/bin/env bash
# 맥미니에서 학습한 모델을 LXC 컨테이너 볼륨 경로로 전송한다.
# 사용법: bash scripts/deploy_model.sh [LXC_HOST] [LXC_MODELS_PATH]
# 사용법: bash scripts/deploy_model.sh [lgbm|mlx]
#
# 예시:
# bash scripts/deploy_model.sh 10.1.10.28 /path/to/cointrader/models
# bash scripts/deploy_model.sh root@10.1.10.28 /root/cointrader/models
# bash scripts/deploy_model.sh # LightGBM (기본값)
# bash scripts/deploy_model.sh mlx # MLX 신경망
set -euo pipefail
LXC_HOST="${1:-root@10.1.10.24}"
LXC_MODELS_PATH="${2:-/root/cointrader/models}"
LOCAL_MODEL="models/lgbm_filter.pkl"
BACKEND="${1:-lgbm}"
LXC_HOST="root@10.1.10.24"
LXC_MODELS_PATH="/root/cointrader/models"
LOCAL_LOG="models/training_log.json"
if [[ ! -f "$LOCAL_MODEL" ]]; then
echo "[오류] 모델 파일 없음: $LOCAL_MODEL"
echo "먼저 python scripts/train_model.py 를 실행하세요."
exit 1
# ── 백엔드별 파일 목록 설정 ──────────────────────────────────────────────────
# mlx: ONNX 파일만 전송 (Linux 서버는 onnxruntime으로 추론)
# lgbm: pkl 파일 전송
RELOAD_CMD="from src.ml_filter import MLFilter; f=MLFilter(); f.reload_model(); print('리로드 완료')"
if [ "$BACKEND" = "mlx" ]; then
LOCAL_FILES=("models/mlx_filter.weights.onnx")
else
LOCAL_FILES=("models/lgbm_filter.pkl")
fi
echo "=== 모델 전송 시작 ==="
echo " 대상: ${LXC_HOST}:${LXC_MODELS_PATH}"
echo " 파일: $LOCAL_MODEL"
# 기존 모델을 prev로 백업 (원격)
ssh "${LXC_HOST}" "
if [ -f '${LXC_MODELS_PATH}/lgbm_filter.pkl' ]; then
cp '${LXC_MODELS_PATH}/lgbm_filter.pkl' '${LXC_MODELS_PATH}/lgbm_filter_prev.pkl'
echo ' 기존 모델 백업 완료'
# ── 파일 존재 확인 ────────────────────────────────────────────────────────────
for f in "${LOCAL_FILES[@]}"; do
if [[ ! -f "$f" ]]; then
echo "[오류] 모델 파일 없음: $f"
exit 1
fi
done
echo "=== 모델 전송 시작 (백엔드: ${BACKEND}) ==="
echo " 대상: ${LXC_HOST}:${LXC_MODELS_PATH}"
# ── 원격 디렉터리 생성 + lgbm 기존 모델 백업 ─────────────────────────────────
ssh "${LXC_HOST}" "
mkdir -p '${LXC_MODELS_PATH}'
if [ '$BACKEND' = 'lgbm' ] && [ -f '${LXC_MODELS_PATH}/lgbm_filter.pkl' ]; then
cp '${LXC_MODELS_PATH}/lgbm_filter.pkl' '${LXC_MODELS_PATH}/lgbm_filter_prev.pkl'
echo ' 기존 lgbm 모델 백업 완료'
fi
"
# 모델 파일 전송 (rsync 우선, 없으면 scp 폴백)
if command -v rsync &>/dev/null && ssh "${LXC_HOST}" "command -v rsync" &>/dev/null; then
rsync -avz --progress \
"$LOCAL_MODEL" \
"${LXC_HOST}:${LXC_MODELS_PATH}/lgbm_filter.pkl"
else
echo " rsync 없음 → scp 사용"
scp "$LOCAL_MODEL" "${LXC_HOST}:${LXC_MODELS_PATH}/lgbm_filter.pkl"
fi
# 학습 로그도 함께 전송 (있을 경우)
if [[ -f "$LOCAL_LOG" ]]; then
# ── 파일 전송 헬퍼 (rsync 우선, scp 폴백) ────────────────────────────────────
_send() {
local src="$1" dst="$2"
echo " 전송: $src → ${LXC_HOST}:$dst"
if command -v rsync &>/dev/null && ssh "${LXC_HOST}" "command -v rsync" &>/dev/null; then
rsync -avz "$LOCAL_LOG" "${LXC_HOST}:${LXC_MODELS_PATH}/training_log.json"
rsync -avz --progress "$src" "${LXC_HOST}:$dst"
else
scp "$LOCAL_LOG" "${LXC_HOST}:${LXC_MODELS_PATH}/training_log.json"
scp "$src" "${LXC_HOST}:$dst"
fi
}
# ── 모델 파일 전송 ────────────────────────────────────────────────────────────
for f in "${LOCAL_FILES[@]}"; do
_send "$f" "${LXC_MODELS_PATH}/$(basename "$f")"
done
# ── 학습 로그 전송 ────────────────────────────────────────────────────────────
if [[ -f "$LOCAL_LOG" ]]; then
_send "$LOCAL_LOG" "${LXC_MODELS_PATH}/training_log.json"
echo " 학습 로그 전송 완료"
fi
echo "=== 전송 완료 ==="
echo ""
# 봇 컨테이너가 실행 중이면 모델 핫리로드, 아니면 건너뜀
# ── 핫리로드 ─────────────────────────────────────────────────────────────────
echo "=== 핫리로드 시도 ==="
if ssh "${LXC_HOST}" "docker inspect -f '{{.State.Running}}' cointrader 2>/dev/null | grep -q true"; then
ssh "${LXC_HOST}" "docker exec cointrader python -c \
\"from src.ml_filter import MLFilter; f=MLFilter(); f.reload_model(); print('리로드 완료')\""
ssh "${LXC_HOST}" "docker exec cointrader python -c \"${RELOAD_CMD}\""
echo "=== 핫리로드 완료 ==="
else
echo " cointrader 컨테이너가 실행 중이 아닙니다. 건너뜁니다."

28
scripts/train_and_deploy.sh
View File

@@ -1,10 +1,10 @@
#!/usr/bin/env bash
# 맥미니에서 전체 학습 파이프라인을 실행하고 LXC로 배포한다.
# 사용법: bash scripts/train_and_deploy.sh [LXC_HOST] [LXC_MODELS_PATH]
# 사용법: bash scripts/train_and_deploy.sh [mlx|lgbm]
#
# 예시:
# bash scripts/train_and_deploy.sh
# bash scripts/train_and_deploy.sh root@10.1.10.24 /root/cointrader/models
# bash scripts/train_and_deploy.sh # LightGBM (기본값)
# bash scripts/train_and_deploy.sh mlx # MLX GPU 학습
set -euo pipefail
@@ -19,37 +19,35 @@ else
echo "경고: 가상환경을 찾을 수 없습니다 ($VENV_PATH). 시스템 Python을 사용합니다." >&2
fi
LXC_HOST="${1:-root@10.1.10.24}"
LXC_MODELS_PATH="${2:-/root/cointrader/models}"
BACKEND="${1:-lgbm}"
cd "$PROJECT_ROOT"
echo "=== [1/3] 데이터 수집 (XRP + BTC + ETH 3심볼) ==="
echo "=== [1/3] 데이터 수집 (XRP + BTC + ETH 3심볼, 1년치) ==="
python scripts/fetch_history.py \
--symbols XRPUSDT BTCUSDT ETHUSDT \
--interval 1m \
--days 90 \
--days 365 \
--output data/xrpusdt_1m.parquet
# 결과: data/combined_1m.parquet (타임스탬프 기준 병합)
echo ""
echo "=== [2/3] 모델 학습 (21개 피처: XRP 13 + BTC/ETH 상관관계 8) ==="
# TRAIN_BACKEND=mlx 로 설정하면 Apple Silicon GPU(Metal)를 사용한다 (기본: lgbm)
BACKEND="${TRAIN_BACKEND:-lgbm}"
DECAY="${TIME_WEIGHT_DECAY:-2.0}"
if [ "$BACKEND" = "mlx" ]; then
echo " 백엔드: MLX (Apple Silicon GPU)"
python scripts/train_mlx_model.py --data data/combined_1m.parquet
echo " 백엔드: MLX (Apple Silicon GPU), decay=${DECAY}"
python scripts/train_mlx_model.py --data data/combined_1m.parquet --decay "$DECAY"
else
echo " 백엔드: LightGBM (CPU)"
python scripts/train_model.py --data data/combined_1m.parquet
echo " 백엔드: LightGBM (CPU), decay=${DECAY}"
python scripts/train_model.py --data data/combined_1m.parquet --decay "$DECAY"
fi
echo ""
echo "=== [3/3] LXC 배포 ==="
bash scripts/deploy_model.sh "$LXC_HOST" "$LXC_MODELS_PATH"
bash scripts/deploy_model.sh "$BACKEND"
echo ""
echo "=== 전체 파이프라인 완료 ==="
echo ""
echo "봇 재시작이 필요하면:"
echo " ssh ${LXC_HOST} 'cd /root/cointrader && docker compose restart cointrader'"
echo " ssh root@10.1.10.24 'cd /root/cointrader && docker compose restart cointrader'"

33
scripts/train_mlx_model.py
View File

@@ -25,14 +25,14 @@ MLX_MODEL_PATH = Path("models/mlx_filter.weights")
LOG_PATH = Path("models/training_log.json")
def train_mlx(data_path: str) -> float:
def train_mlx(data_path: str, time_weight_decay: float = 2.0) -> float:
print(f"데이터 로드: {data_path}")
df = pd.read_parquet(data_path)
print(f"캔들 수: {len(df)}")
print("\n데이터셋 생성 중...")
t0 = time.perf_counter()
dataset = generate_dataset_vectorized(df)
dataset = generate_dataset_vectorized(df, time_weight_decay=time_weight_decay)
t1 = time.perf_counter()
print(f"데이터셋 생성 완료: {t1 - t0:.1f}초, {len(dataset)}개 샘플")
@@ -46,10 +46,30 @@ def train_mlx(data_path: str) -> float:
X = dataset[FEATURE_COLS]
y = dataset["label"]
w = dataset["sample_weight"].values
split = int(len(X) * 0.8)
X_train, X_val = X.iloc[:split], X.iloc[split:]
y_train, y_val = y.iloc[:split], y.iloc[split:]
w_train = w[:split]
# --- 클래스 불균형 처리: 언더샘플링 (가중치 인덱스 보존) ---
pos_idx = np.where(y_train == 1)[0]
neg_idx = np.where(y_train == 0)[0]
if len(neg_idx) > len(pos_idx):
np.random.seed(42)
neg_idx = np.random.choice(neg_idx, size=len(pos_idx), replace=False)
balanced_idx = np.concatenate([pos_idx, neg_idx])
np.random.shuffle(balanced_idx)
X_train = X_train.iloc[balanced_idx]
y_train = y_train.iloc[balanced_idx]
w_train = w_train[balanced_idx]
print(f"\n언더샘플링 적용 후 학습 데이터: {len(X_train)}개 (양성={y_train.sum()}, 음성={(y_train==0).sum()})")
# --------------------------------------
print("\nMLX 신경망 학습 시작 (GPU)...")
t2 = time.perf_counter()
@@ -60,7 +80,7 @@ def train_mlx(data_path: str) -> float:
epochs=100,
batch_size=256,
)
model.fit(X_train, y_train)
model.fit(X_train, y_train, sample_weight=w_train)
t3 = time.perf_counter()
print(f"학습 완료: {t3 - t2:.1f}초")
@@ -83,6 +103,7 @@ def train_mlx(data_path: str) -> float:
"auc": round(auc, 4),
"samples": len(dataset),
"train_sec": round(t3 - t2, 1),
"time_weight_decay": time_weight_decay,
"model_path": str(MLX_MODEL_PATH),
})
with open(LOG_PATH, "w") as f:
@@ -94,8 +115,12 @@ def train_mlx(data_path: str) -> float:
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--data", default="data/xrpusdt_1m.parquet")
parser.add_argument(
"--decay", type=float, default=2.0,
help="시간 가중치 감쇠 강도 (0=균등, 2.0=최신이 ~7.4배 높음)",
)
args = parser.parse_args()
train_mlx(args.data)
train_mlx(args.data, time_weight_decay=args.decay)
if __name__ == "__main__":

33
scripts/train_model.py
View File

@@ -146,7 +146,7 @@ def generate_dataset(df: pd.DataFrame, n_jobs: int | None = None) -> pd.DataFram
return pd.DataFrame(rows)
def train(data_path: str):
def train(data_path: str, time_weight_decay: 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)}")
@@ -169,7 +169,7 @@ def train(data_path: str):
df = df_raw[base_cols].copy()
print("데이터셋 생성 중...")
dataset = generate_dataset_vectorized(df, btc_df=btc_df, eth_df=eth_df)
dataset = generate_dataset_vectorized(df, btc_df=btc_df, eth_df=eth_df, time_weight_decay=time_weight_decay)
if dataset.empty or "label" not in dataset.columns:
raise ValueError(f"데이터셋 생성 실패: 샘플 0개. 위 오류 메시지를 확인하세요.")
@@ -183,10 +183,30 @@ def train(data_path: str):
print(f"사용 피처: {len(actual_feature_cols)}개 {actual_feature_cols}")
X = dataset[actual_feature_cols]
y = dataset["label"]
w = dataset["sample_weight"].values
split = int(len(X) * 0.8)
X_train, X_val = X.iloc[:split], X.iloc[split:]
y_train, y_val = y.iloc[:split], y.iloc[split:]
w_train = w[:split]
# --- 클래스 불균형 처리: 언더샘플링 (가중치 인덱스 보존) ---
pos_idx = np.where(y_train == 1)[0]
neg_idx = np.where(y_train == 0)[0]
if len(neg_idx) > len(pos_idx):
np.random.seed(42)
neg_idx = np.random.choice(neg_idx, size=len(pos_idx), replace=False)
balanced_idx = np.concatenate([pos_idx, neg_idx])
np.random.shuffle(balanced_idx)
X_train = X_train.iloc[balanced_idx]
y_train = y_train.iloc[balanced_idx]
w_train = w_train[balanced_idx]
print(f"\n언더샘플링 적용 후 학습 데이터: {len(X_train)}개 (양성={y_train.sum()}, 음성={(y_train==0).sum()})")
# --------------------------------------
model = lgb.LGBMClassifier(
n_estimators=300,
@@ -201,6 +221,7 @@ def train(data_path: str):
)
model.fit(
X_train, y_train,
sample_weight=w_train,
eval_set=[(X_val, y_val)],
callbacks=[lgb.early_stopping(30, verbose=False), lgb.log_evaluation(50)],
)
@@ -225,9 +246,11 @@ def train(data_path: str):
log = json.load(f)
log.append({
"date": datetime.now().isoformat(),
"backend": "lgbm",
"auc": round(auc, 4),
"samples": len(dataset),
"features": len(actual_feature_cols),
"time_weight_decay": time_weight_decay,
"model_path": str(MODEL_PATH),
})
with open(LOG_PATH, "w") as f:
@@ -239,8 +262,12 @@ def train(data_path: str):
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--data", default="data/xrpusdt_1m.parquet")
parser.add_argument(
"--decay", type=float, default=2.0,
help="시간 가중치 감쇠 강도 (0=균등, 2.0=최신이 ~7.4배 높음)",
)
args = parser.parse_args()
train(args.data)
train(args.data, time_weight_decay=args.decay)
if __name__ == "__main__":