fithia2/scripts/v11_experiments.py

#!/usr/bin/env python3
"""V11 experiment runner: structural improvements over V10 champion.

Tests VIX regime filter, entropy scoring, and atr_ratio scoring for gainers_leader.
All experiments use V10 (orb_gainers_v10.yaml) as baseline.

Usage:
    python -u scripts/v11_experiments.py [--days N]
"""
import subprocess
import sys
import tempfile
import yaml
from pathlib import Path

BASE_CONFIG = "configs/intraday/strategies/orb_gainers_v10.yaml"
DAYS = 400


def load_base():
    with open(BASE_CONFIG) as f:
        return yaml.safe_load(f)


def run_experiment(name: str, overrides: dict, days: int = DAYS) -> dict | None:
    """Run a single backtest with config overrides and return parsed metrics."""
    cfg = load_base()
    for key, val in overrides.items():
        cfg["orb_strategy"][key] = val

    # Write temp config
    with tempfile.NamedTemporaryFile(
        mode="w", suffix=".yaml", prefix="v11_", delete=False, dir="/tmp"
    ) as f:
        yaml.dump(cfg, f, default_flow_style=False)
        tmp_path = f.name

    cmd = [
        sys.executable, "-u", "-m", "apps.intraday_bt.run",
        "--config", tmp_path,
        "--days", str(days),
    ]

    print(f"\n{'='*70}")
    print(f"  {name}")
    print(f"  Overrides: {overrides or '(baseline)'}")
    print(f"{'='*70}", flush=True)

    try:
        result = subprocess.run(cmd, capture_output=True, text=True, timeout=600)
    except subprocess.TimeoutExpired:
        print(f"  TIMEOUT after 600s")
        return None
    finally:
        Path(tmp_path).unlink(missing_ok=True)

    if result.returncode != 0:
        print(f"  FAILED (rc={result.returncode})")
        if result.stderr:
            print(f"  stderr: {result.stderr[-500:]}")
        return None

    output = result.stdout
    metrics = {}
    for line in output.split("\n"):
        line = line.strip()
        if "Total return" in line and "%" in line:
            try:
                val = line.split("│")[-2].strip().replace("%", "").replace("+", "")
                metrics["total_return"] = float(val)
            except (ValueError, IndexError):
                pass
        elif "Max drawdown" in line and "%" in line:
            try:
                val = line.split("│")[-2].strip().replace("%", "").replace("+", "")
                metrics["max_drawdown"] = float(val)
            except (ValueError, IndexError):
                pass
        elif "Sharpe ratio" in line:
            try:
                val = line.split("│")[-2].strip()
                metrics["sharpe"] = float(val)
            except (ValueError, IndexError):
                pass
        elif "Total trades" in line:
            try:
                val = line.split("│")[-2].strip()
                metrics["trades"] = int(val)
            except (ValueError, IndexError):
                pass
        elif "Win rate" in line and "%" in line:
            try:
                val = line.split("│")[-2].strip().replace("%", "").replace("+", "")
                metrics["win_rate"] = float(val)
            except (ValueError, IndexError):
                pass
        elif "Profit factor" in line:
            try:
                val = line.split("│")[-2].strip()
                metrics["profit_factor"] = float(val)
            except (ValueError, IndexError):
                pass

    if not metrics:
        print("  WARNING: Could not parse metrics from output")
        for line in output.split("\n")[-30:]:
            print(f"    {line}")
        return None

    print(f"  => return={metrics.get('total_return', '?'):+.2f}%, "
          f"DD={metrics.get('max_drawdown', '?'):.2f}%, "
          f"Sharpe={metrics.get('sharpe', '?'):.2f}, "
          f"trades={metrics.get('trades', '?')}, "
          f"WR={metrics.get('win_rate', '?'):.1f}%, "
          f"PF={metrics.get('profit_factor', '?'):.3f}", flush=True)
    return metrics


def main():
    days = DAYS
    if "--days" in sys.argv:
        idx = sys.argv.index("--days")
        days = int(sys.argv[idx + 1])

    results = {}

    # V10 baseline (no overrides)
    results["V10_baseline"] = run_experiment("V10 Baseline (control)", {}, days)

    # === VIX experiments ===
    results["V11a_max_vix_30"] = run_experiment(
        "V11a: max_vix=30 (skip VIX>30 days)", {"max_vix": 30}, days
    )
    results["V11a2_max_vix_25"] = run_experiment(
        "V11a2: max_vix=25 (more aggressive)", {"max_vix": 25}, days
    )
    results["V11b_vix_scale"] = run_experiment(
        "V11b: VIX size scaling 15→30, min=0.5",
        {"vix_size_scale_low": 15, "vix_size_scale_high": 30, "vix_size_scale_min": 0.5},
        days,
    )
    results["V11b2_vix_scale_wide"] = run_experiment(
        "V11b2: VIX size scaling 20→35, min=0.5",
        {"vix_size_scale_low": 20, "vix_size_scale_high": 35, "vix_size_scale_min": 0.5},
        days,
    )

    # === Entropy experiments (now unlocked for gainers_leader) ===
    results["V11c_entropy_neg10"] = run_experiment(
        "V11c: weight_entropy=-0.10 (prefer lower entropy)",
        {"weight_entropy": -0.10}, days,
    )
    results["V11c2_entropy_neg05"] = run_experiment(
        "V11c2: weight_entropy=-0.05",
        {"weight_entropy": -0.05}, days,
    )
    results["V11c3_entropy_pos10"] = run_experiment(
        "V11c3: weight_entropy=+0.10 (prefer higher entropy)",
        {"weight_entropy": 0.10}, days,
    )

    # === ATR ratio experiments (now unlocked for gainers_leader) ===
    results["V11d_atr_ratio_pos10"] = run_experiment(
        "V11d: weight_atr_ratio=+0.10 (prefer expanding vol)",
        {"weight_atr_ratio": 0.10}, days,
    )
    results["V11d2_atr_ratio_neg10"] = run_experiment(
        "V11d2: weight_atr_ratio=-0.10 (prefer compressed vol)",
        {"weight_atr_ratio": -0.10}, days,
    )

    # === Combinations ===
    results["V11e_vix30_entropy"] = run_experiment(
        "V11e: max_vix=30 + entropy=-0.10",
        {"max_vix": 30, "weight_entropy": -0.10}, days,
    )
    results["V11f_vix30_atr_ratio"] = run_experiment(
        "V11f: max_vix=30 + atr_ratio=+0.10",
        {"max_vix": 30, "weight_atr_ratio": 0.10}, days,
    )
    results["V11g_combined"] = run_experiment(
        "V11g: VIX scale 20→35 + entropy=-0.05 + atr_ratio=+0.10",
        {
            "vix_size_scale_low": 20, "vix_size_scale_high": 35, "vix_size_scale_min": 0.5,
            "weight_entropy": -0.05, "weight_atr_ratio": 0.10,
        },
        days,
    )

    # === Summary table ===
    print(f"\n\n{'='*90}")
    print("  V11 EXPERIMENT RESULTS SUMMARY")
    print(f"{'='*90}")
    print(f"{'Experiment':<30} {'Return':>9} {'DD':>9} {'Sharpe':>7} {'Trades':>7} {'WR':>7} {'PF':>7}")
    print(f"{'-'*30} {'-'*9} {'-'*9} {'-'*7} {'-'*7} {'-'*7} {'-'*7}")

    baseline_ret = (results.get("V10_baseline") or {}).get("total_return", 0)
    for name, m in results.items():
        if m is None:
            print(f"{name:<30} {'FAILED':>9}")
            continue
        ret = m.get("total_return", 0)
        delta = ret - baseline_ret
        ret_str = f"{ret:+.2f}%"
        dd = f"{m.get('max_drawdown', 0):.2f}%"
        sh = f"{m.get('sharpe', 0):.2f}"
        tr = f"{m.get('trades', 0)}"
        wr = f"{m.get('win_rate', 0):.1f}%"
        pf = f"{m.get('profit_factor', 0):.3f}"
        marker = " <== BASE" if name == "V10_baseline" else (f" ({delta:+.2f}pp)" if delta != 0 else "")
        print(f"{name:<30} {ret_str:>9} {dd:>9} {sh:>7} {tr:>7} {wr:>7} {pf:>7}{marker}")


if __name__ == "__main__":
    main()