File size: 6,676 Bytes
e365f22 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 | """
threshold_optimizer.py — Post-training threshold calibration tool.
Run this standalone to re-optimize the probability threshold on new data
WITHOUT retraining the model. Useful for:
- Adapting to regime changes without full retraining
- Testing different optimization objectives
- Out-of-sample threshold validation
The threshold search maximizes expectancy or Sharpe over a held-out dataset.
Usage:
python threshold_optimizer.py --symbols BTC-USDT ETH-USDT --bars 200
python threshold_optimizer.py --objective sharpe
"""
import argparse
import json
import logging
import sys
from pathlib import Path
import numpy as np
import pandas as pd
import matplotlib
matplotlib.use("Agg") # non-interactive backend
import matplotlib.pyplot as plt
sys.path.insert(0, str(Path(__file__).parent))
from ml_config import (
THRESHOLD_PATH,
THRESHOLD_MIN,
THRESHOLD_MAX,
THRESHOLD_STEPS,
THRESHOLD_OBJECTIVE,
TARGET_RR,
ROUND_TRIP_COST,
FEATURE_COLUMNS,
ML_DIR,
)
from ml_filter import TradeFilter
from feature_builder import build_feature_dict, validate_features
from train import build_dataset
logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s")
def compute_threshold_curve(
probs: np.ndarray,
y_true: np.ndarray,
rr: float = TARGET_RR,
cost: float = ROUND_TRIP_COST,
) -> pd.DataFrame:
"""
Sweep threshold grid and compute metrics at each threshold.
Returns DataFrame for analysis and plotting.
"""
thresholds = np.linspace(THRESHOLD_MIN, THRESHOLD_MAX, THRESHOLD_STEPS)
records = []
for t in thresholds:
mask = probs >= t
n = int(mask.sum())
if n < 5:
records.append({
"threshold": t, "n_trades": n,
"win_rate": np.nan, "expectancy": np.nan,
"sharpe": np.nan, "precision": np.nan,
"coverage": 0.0,
})
continue
y_f = y_true[mask]
wr = float(y_f.mean())
exp = wr * rr - (1 - wr) * 1.0 - cost
pnl = np.where(y_f == 1, rr, -1.0) - cost
sh = (pnl.mean() / pnl.std() * np.sqrt(252)) if pnl.std() > 1e-9 else 0.0
cov = n / len(y_true)
records.append({
"threshold": round(t, 4),
"n_trades": n,
"win_rate": round(wr, 4),
"expectancy": round(exp, 4),
"sharpe": round(sh, 4),
"precision": round(wr, 4),
"coverage": round(cov, 4),
})
return pd.DataFrame(records)
def find_optimal_threshold(
curve: pd.DataFrame,
objective: str = THRESHOLD_OBJECTIVE,
min_trades: int = 20,
) -> float:
valid = curve[curve["n_trades"] >= min_trades].dropna(subset=[objective])
if valid.empty:
logger.warning("No valid threshold found — using default 0.55")
return 0.55
best_row = valid.loc[valid[objective].idxmax()]
return float(best_row["threshold"])
def plot_threshold_curves(curve: pd.DataFrame, optimal: float, save_path: Path):
fig, axes = plt.subplots(2, 2, figsize=(12, 8))
fig.suptitle("Threshold Optimization", fontsize=14, fontweight="bold")
metrics = ["expectancy", "sharpe", "win_rate", "n_trades"]
titles = ["Expectancy per Trade", "Annualized Sharpe", "Win Rate", "# Trades"]
for ax, metric, title in zip(axes.flatten(), metrics, titles):
valid = curve.dropna(subset=[metric])
ax.plot(valid["threshold"], valid[metric], lw=2, color="#1a6bff")
ax.axvline(optimal, color="orange", linestyle="--", lw=1.5, label=f"Optimal={optimal:.3f}")
ax.axhline(0, color="gray", linestyle=":", lw=0.8)
ax.set_title(title, fontsize=11)
ax.set_xlabel("Threshold")
ax.legend(fontsize=9)
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig(save_path, dpi=120, bbox_inches="tight")
plt.close()
logger.info(f"Threshold curve plot saved → {save_path}")
def main(args):
trade_filter = TradeFilter.load_or_none()
if trade_filter is None:
logger.error("No trained model found. Run train.py first.")
sys.exit(1)
symbols = args.symbols or ["BTC-USDT", "ETH-USDT", "SOL-USDT", "BNB-USDT"]
dataset = build_dataset(symbols, bars=args.bars)
X = dataset[FEATURE_COLUMNS].values.astype(np.float64)
y = dataset["label"].values.astype(np.int32)
feature_dicts = [
{k: float(row[k]) for k in FEATURE_COLUMNS}
for _, row in dataset[FEATURE_COLUMNS].iterrows()
]
probs = trade_filter.predict_batch(feature_dicts)
logger.info(f"Generated {len(probs)} predictions | mean_prob={probs.mean():.4f}")
curve = compute_threshold_curve(probs, y)
optimal = find_optimal_threshold(curve, objective=args.objective)
best_row = curve[curve["threshold"].round(4) == round(optimal, 4)].iloc[0]
logger.info(f"\n=== THRESHOLD OPTIMIZATION RESULT ===")
logger.info(f" Objective: {args.objective}")
logger.info(f" Optimal threshold: {optimal:.4f}")
logger.info(f" Win rate: {best_row['win_rate']:.4f}")
logger.info(f" Expectancy: {best_row['expectancy']:.4f}")
logger.info(f" Sharpe: {best_row['sharpe']:.4f}")
logger.info(f" # Trades: {int(best_row['n_trades'])}")
logger.info(f" Coverage: {best_row['coverage']:.2%}")
# Update threshold file
ML_DIR.mkdir(parents=True, exist_ok=True)
thresh_data = {
"threshold": optimal,
"objective": args.objective,
"win_rate_at_threshold": float(best_row["win_rate"]),
"expectancy_at_threshold": float(best_row["expectancy"]),
"sharpe_at_threshold": float(best_row["sharpe"]),
"n_trades_at_threshold": int(best_row["n_trades"]),
}
with open(THRESHOLD_PATH, "w") as f:
json.dump(thresh_data, f, indent=2)
logger.info(f"Threshold updated → {THRESHOLD_PATH}")
# Save curve CSV
curve_path = ML_DIR / "threshold_curve.csv"
curve.to_csv(curve_path, index=False)
# Plot
plot_path = ML_DIR / "threshold_curve.png"
try:
plot_threshold_curves(curve, optimal, plot_path)
except Exception as e:
logger.warning(f"Plot failed: {e}")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Optimize probability threshold")
parser.add_argument("--symbols", nargs="+", default=None)
parser.add_argument("--bars", type=int, default=200)
parser.add_argument("--objective", choices=["expectancy", "sharpe", "win_rate"], default=THRESHOLD_OBJECTIVE)
args = parser.parse_args()
main(args)
|