scripts/sweep.py --target signal finds the optimal inference and execution parameters by running many OOS backtests across a parameter grid. No model retraining — it only changes config.json and ml_config.json values that affect signal filtering and risk sizing.
Developer/operator lane only. Regular users should run onboarding, select profile 1-5, pull approved model revisions from R2, and run trading.
Usage
# Sweep confidence threshold + min_probability_diff
python scripts/sweep.py --target signal --mode confidence
# Sweep TP/SL ATR multipliers
python scripts/sweep.py --target signal --mode sltp
# Sweep risk_percent + max_consecutive_losses
python scripts/sweep.py --target signal --mode risk
# All modes combined (~68 configs, ~20 min)
python scripts/sweep.py --target signal --mode full --output results/sweep_$(date +%Y%m%d).csv
# Single-axis custom sweep
python scripts/sweep.py --target signal --mode custom \
--param dynamic_sltp.tp_atr_multiplier \
--values 0.7 0.8 0.9 1.0 1.2
Sweep modes
confidence
Tests combinations of confidence_threshold (0.50–0.65) and min_probability_diff (0.05–0.15).
| Parameter | Path in config | Sweep range |
|---|
confidence_threshold | ml_config.json → prediction | 0.50, 0.52, 0.55, 0.58, 0.60, 0.62, 0.65 |
min_probability_diff | ml_config.json → prediction | 0.05, 0.08, 0.10, 0.12, 0.15 |
sltp
Tests ATR multiplier combinations for SL and TP.
| Parameter | Path in config | Sweep range |
|---|
sl_atr_multiplier | config.json → dynamic_sltp | 0.6, 0.7, 0.8, 0.9, 1.0 |
tp_atr_multiplier | config.json → dynamic_sltp | 0.6, 0.7, 0.8, 1.0, 1.2, 1.5 |
risk
Tests risk sizing and circuit breaker combinations.
| Parameter | Path in config | Sweep range |
|---|
risk_percent | config.json → dynamic_position_sizing | 1, 2, 3, 4, 5, 6, 8 |
max_consecutive_losses | config.json | 3, 4, 5, 6, 8, 10 |
custom
Sweep any single parameter with explicit values:
python scripts/sweep.py --target signal --mode custom \
--param ml_config.prediction.confidence_threshold \
--values 0.55 0.58 0.60 0.62 0.65
--param flag accepts dot-notation paths into config.json or ml_config.json.
CLI flags
| Flag | Default | Description |
|---|
--mode | required | confidence, sltp, risk, full, or custom |
--param | — | Dot-notation key for --mode custom |
--values | — | Space-separated values for --mode custom |
--output | auto | CSV file path for results |
--balance | 500 | Backtest starting balance ($) |
--leverage | 500 | Backtest leverage |
--spread | 16.95 | Spread in $ (16.95 = VT Markets, 3.0 = IC Markets) |
Reading results
Results are printed as a ranked table and saved to CSV:
Rank WR PF MaxDD Sharpe Score conf prob_diff
1 78.5% 2.43 1.8% 32.99 21.34 0.60 0.10
2 75.9% 2.19 2.2% 27.46 17.74 0.58 0.10
3 73.8% 2.05 2.5% 24.20 15.18 0.56 0.12
...
Score = WR × PF / sqrt(MaxDD)
# View saved CSV
cat results/sweep_signal_*.csv | column -t -s ','
Applying the best result
The script does not automatically apply the best config. Apply manually:
# Edit ml_config.json
python3 -c "
import json
ml = json.load(open('ml_config.json'))
ml['prediction']['confidence_threshold'] = 0.60
ml['prediction']['min_probability_diff'] = 0.10
with open('ml_config.json', 'w') as f: json.dump(ml, f, indent=2)
print('Applied')
"
# Verify with a final OOS backtest
python backtest_config.py \
--balance 500 --no-swap --leverage 500 \
--spread 16.95 --oos-only --no-chart
risk-tuner Claude agent for fully automated sweeping + application.
