Community Scripts

# ╔══════════════════════════════════════════════════════════════╗
# ║                  STRATEGY REQUEST LOG                       ║
# ╚══════════════════════════════════════════════════════════════╝
# Generated        : 2026-05-06 01:41:21
# Model            : Gradient Boosting
# Feature Eng.     : SMA (20,50,200), BB (20,2.0), RSI 14, MACD (12,26,9), ATR 14 + Auto-add features: ON
# Signal / Entry   : Enter when model confidence > threshold; exit on opposite signal or SL/TP
# Optimization     : Maximize risk-adjusted return
# Risk Mgmt        : Stop loss 0.5%, Take profit 1.0%
# Risk Filter      : —
# ══════════════════════════════════════════════════════════════

# ============================================================
# SECTION 0 — IMPORTS & CONSTANTS
import numpy as np
import pandas as pd

DATA_PATH = "/root/Desktop/QuantifyMe/data/ohlc/GBPUSD_15min.parquet"
START_DATE = "2025-04-24"
END_DATE   = "2026-04-24"
VALIDATION_DATE = ""
TRAIN_SPLIT = 0.7

# SECTION 1 — FEATURE ENGINEERING
def feature_engineering(df, close, open_, high, low):

    # ── SMA 20, 50, 200 + distance from close ──────────────────────────────
    for period in [20, 50, 200]:
        sma = close.rolling(period).mean()
        df[f"sma_{period}"] = sma
        df[f"dm_sma_{period}"] = (close - sma) / sma

    # ── Bollinger Bands (20, 2) ─────────────────────────────────────────────
    bb_mid   = close.rolling(20).mean()
    bb_std   = close.rolling(20).std(ddof=0)
    bb_upper = bb_mid + 2.0 * bb_std
    bb_lower = bb_mid - 2.0 * bb_std
    df["bb_mid"]   = bb_mid
    df["bb_upper"] = bb_upper
    df["bb_lower"] = bb_lower
    df["bb_width"] = (bb_upper - bb_lower) / bb_mid
    bb_range       = bb_upper - bb_lower
    df["bb_pct"]   = np.where(bb_range != 0, (close - bb_lower) / bb_range, 0.5)

    # ── RSI 14 ─────────────────────────────────────────────────────────────
    delta     = close.diff()
    gain      = delta.clip(lower=0)
    loss      = (-delta).clip(lower=0)
    avg_gain  = gain.ewm(com=13, min_periods=14, adjust=False).mean()
    avg_loss  = loss.ewm(com=13, min_periods=14, adjust=False).mean()
    rs        = np.where(avg_loss != 0, avg_gain / avg_loss, 100.0)
    df["rsi_14"] = 100.0 - (100.0 / (1.0 + rs))

    # ── MACD (12, 26, 9) ───────────────────────────────────────────────────
    ema_12          = close.ewm(span=12, adjust=False).mean()
    ema_26          = close.ewm(span=26, adjust=False).mean()
    macd_line       = ema_12 - ema_26
    signal_line     = macd_line.ewm(span=9, adjust=False).mean()
    df["macd_line"] = macd_line
    df["macd_signal"] = signal_line
    df["macd_hist"] = macd_line - signal_line

    # ── ATR 14 + NATR ──────────────────────────────────────────────────────
    prev_close  = close.shift(1)
    tr          = pd.concat([
        high - low,
        (high - prev_close).abs(),
        (low  - prev_close).abs()
    ], axis=1).max(axis=1)
    atr         = tr.ewm(com=13, min_periods=14, adjust=False).mean()
    df["atr_14"] = atr
    df["natr"]   = np.where(close != 0, atr / close, 0.0)

    # ── Price momentum (returns over multiple horizons) ────────────────────
    for lag in [1, 4, 8, 16]:
        df[f"ret_{lag}"] = close.pct_change(lag)

    # ── Candle body & wick features ────────────────────────────────────────
    body             = (close - open_).abs()
    candle_range     = (high - low).replace(0, np.nan)
    df["body_ratio"] = body / candle_range
    df["upper_wick"] = np.where(
        candle_range.notna(),
        (high - close.combine(open_, max)) / candle_range,
        0.0
    )
    df["lower_wick"] = np.where(
        candle_range.notna(),
        (close.combine(open_, min) - low) / candle_range,
        0.0
    )
    df["candle_dir"] = np.where(close >= open_, 1.0, -1.0)

    # ── Volume proxy: normalised candle range ──────────────────────────────
    rolling_range    = candle_range.rolling(20).mean()
    df["norm_range"] = np.where(
        rolling_range != 0,
        (high - low) / rolling_range,
        1.0
    )

    # ── RSI derived features ───────────────────────────────────────────────
    df["rsi_ob"]    = np.where(df["rsi_14"] > 70, 1.0, 0.0)
    df["rsi_os"]    = np.where(df["rsi_14"] < 30, 1.0, 0.0)
    df["rsi_mid"]   = df["rsi_14"] - 50.0
    df["rsi_slope"] = df["rsi_14"].diff(3)

    # ── MACD histogram slope ───────────────────────────────────────────────
    df["macd_hist_slope"] = df["macd_hist"].diff(2)
    df["macd_cross"]      = np.where(
        (df["macd_hist"] > 0) & (df["macd_hist"].shift(1) <= 0), 1.0,
        np.where(
            (df["macd_hist"] < 0) & (df["macd_hist"].shift(1) >= 0), -1.0,
            0.0
        )
    )

    # ── Bollinger squeeze (low volatility precursor) ───────────────────────
    bb_width_ma         = df["bb_width"].rolling(20).mean()
    df["bb_squeeze"]    = np.where(df["bb_width"] < bb_width_ma, 1.0, 0.0)

    # ── SMA slope features ─────────────────────────────────────────────────
    df["sma_20_slope"] = df["sma_20"].pct_change(4)
    df["sma_50_slope"] = df["sma_50"].pct_change(8)

    # ── Cross-SMA alignment (trend structure) ──────────────────────────────
    df["sma20_above_50"]  = np.where(df["sma_20"] > df["sma_50"],  1.0, 0.0)
    df["sma50_above_200"] = np.where(df["sma_50"] > df["sma_200"], 1.0, 0.0)
    df["close_above_200"] = np.where(close > df["sma_200"],        1.0, 0.0)

    # ── Lagged close returns as additional features ────────────────────────
    for lag in [1, 2, 3]:
        df[f"close_lag_{lag}"] = close.shift(lag)

    # ── Rolling volatility (std of returns) ────────────────────────────────
    df["vol_10"]  = close.pct_change().rolling(10).std()
    df["vol_20"]  = close.pct_change().rolling(20).std()
    df["vol_ratio"] = np.where(
        df["vol_20"] != 0,
        df["vol_10"] / df["vol_20"],
        1.0
    )

    # ── Hour-of-day (London/NY session proxy) ──────────────────────────────
    if hasattr(df.index, "hour"):
        df["hour_sin"] = np.sin(2 * np.pi * df.index.hour / 24.0)
        df["hour_cos"] = np.cos(2 * np.pi * df.index.hour / 24.0)
    else:
        df["hour_sin"] = 0.0
        df["hour_cos"] = 1.0

    # ── Fill NaN from warm-up periods ──────────────────────────────────────
    df = df.bfill().ffill()

    return df


# SECTION 2 — STRATEGY CONFIG
def strategy_config():
    return {
        "title":            "GBP/USD Gradient Boosting Trend + Mean-Reversion",
        "model_type":       "GradientBoostingClassifier",
        "model_params":     {
            "n_estimators":     400,
            "max_depth":        4,
            "learning_rate":    0.04,
            "subsample":        0.8,
            "min_samples_leaf": 20,
            "max_features":     "sqrt",
            "validation_fraction": 0.1,
            "n_iter_no_change": 30,
            "tol":              1e-4,
            "random_state":     42,
        },
        "signal_threshold": 0.55,
        "direction":        "both",
        "stop_loss":        0.005,
        "take_profit":      0.010,
        "cooldown":         0,
        "max_positions":    1,
        "on_opposite":      "reverse",
        "session_filter":   [6, 20],
        "min_atr":          0.0002,
        "trend_filter":     None,
        "target_horizon":   4,
        "objective": (
            "Maximize risk-adjusted return (Sharpe/Calmar) on GBP/USD 15-min. "
            "GradientBoostingClassifier with 400 shallow trees (depth 4) and a "
            "conservative learning rate of 0.04 avoids overfitting while capturing "
            "non-linear interactions between trend (SMA alignment, slope), momentum "
            "(MACD histogram, RSI), and volatility (ATR, BB squeeze) features. "
            "SL=0.5% / TP=1.0% gives 1:2 R/R. Session filter 06-20 UTC covers "
            "London open through NY close where GBP/USD liquidity is highest. "
            "min_atr filter avoids flat/illiquid bars."
        ),
        "notes": (
            "Features include multi-period SMA distances, Bollinger Band pct/width, "
            "RSI with overbought/oversold flags, MACD histogram slope and crossover, "
            "ATR-normalised volatility, candle body/wick ratios, rolling vol ratio, "
            "and hour-of-day cyclical encoding. target_horizon=4 (1-hour forward) "
            "balances signal frequency against predictability at 15-min resolution."
        ),
    }

# ╔══════════════════════════════════════════════════════════════╗
# ║                  STRATEGY REQUEST LOG                       ║
# ╚══════════════════════════════════════════════════════════════╝
# Generated        : 2026-05-06 01:53:28
# Model            : Gradient Boosting
# Feature Eng.     : BB (20,2.0), ATR 14 + Auto-add features: ON
# Signal / Entry   : Enter when model confidence > threshold; exit on opposite signal or SL/TP
# Optimization     : Maximize risk-adjusted return
# Risk Mgmt        : Stop loss 0.5%, Take profit 1.0%
# Risk Filter      : —
# ══════════════════════════════════════════════════════════════

# ============================================================
# Bollinger Bands Squeeze Breakout — GBP/USD 15-min
# ============================================================

# SECTION 0 — IMPORTS & CONSTANTS
import numpy as np
import pandas as pd

DATA_PATH = "/root/Desktop/QuantifyMe/data/ohlc/GBPUSD_15min.parquet"
START_DATE = "2025-04-24"
END_DATE   = "2026-04-24"
VALIDATION_DATE = ""
TRAIN_SPLIT = 0.7

# SECTION 1 — FEATURE ENGINEERING
def feature_engineering(df, close, open_, high, low):
    # ── Bollinger Bands (20, 2) ──────────────────────────────────────────────
    bb_period = 20
    bb_std    = 2.0

    bb_mid   = close.rolling(bb_period).mean()
    bb_sigma = close.rolling(bb_period).std(ddof=0)
    bb_upper = bb_mid + bb_std * bb_sigma
    bb_lower = bb_mid - bb_std * bb_sigma

    bb_width = (bb_upper - bb_lower) / bb_mid
    bb_pct   = (close - bb_lower) / (bb_upper - bb_lower)

    df["bb_mid"]   = bb_mid
    df["bb_upper"] = bb_upper
    df["bb_lower"] = bb_lower
    df["bb_width"] = bb_width
    df["bb_pct"]   = bb_pct

    # ── ATR (14) & NATR ─────────────────────────────────────────────────────
    atr_period = 14
    prev_close = close.shift(1)
    tr = pd.concat([
        high - low,
        (high - prev_close).abs(),
        (low  - prev_close).abs()
    ], axis=1).max(axis=1)
    atr  = tr.ewm(span=atr_period, min_periods=atr_period, adjust=False).mean()
    natr = atr / close

    df["atr"]  = atr
    df["natr"] = natr

    # ── Squeeze detection ────────────────────────────────────────────────────
    # Squeeze = BB width is in the bottom quartile over a 50-bar lookback
    bb_width_min = bb_width.rolling(50).min()
    bb_width_max = bb_width.rolling(50).max()
    bb_width_norm = (bb_width - bb_width_min) / (bb_width_max - bb_width_min + 1e-12)

    df["bb_width_norm"] = bb_width_norm
    df["squeeze"]       = np.where(bb_width_norm < 0.25, 1.0, 0.0)

    # Squeeze released: was in squeeze 1 bar ago, now width is expanding
    bb_width_chg = bb_width.diff()
    df["squeeze_release"] = np.where(
        (df["squeeze"].shift(1) == 1.0) & (bb_width_chg > 0), 1.0, 0.0
    )

    # ── BB width momentum ────────────────────────────────────────────────────
    df["bb_width_chg"]    = bb_width_chg
    df["bb_width_chg_2"]  = bb_width.diff(2)
    df["bb_width_chg_5"]  = bb_width.diff(5)

    # ── Price position relative to bands ─────────────────────────────────────
    df["close_vs_mid"]   = close - bb_mid
    df["close_vs_upper"] = close - bb_upper
    df["close_vs_lower"] = close - bb_lower

    # ── Momentum & returns ───────────────────────────────────────────────────
    df["ret_1"]  = close.pct_change(1)
    df["ret_3"]  = close.pct_change(3)
    df["ret_5"]  = close.pct_change(5)
    df["ret_10"] = close.pct_change(10)
    df["ret_20"] = close.pct_change(20)

    # ── RSI (14) ─────────────────────────────────────────────────────────────
    rsi_period = 14
    delta = close.diff()
    gain  = delta.clip(lower=0)
    loss  = (-delta).clip(lower=0)
    avg_gain = gain.ewm(span=rsi_period, min_periods=rsi_period, adjust=False).mean()
    avg_loss = loss.ewm(span=rsi_period, min_periods=rsi_period, adjust=False).mean()
    rs  = avg_gain / (avg_loss + 1e-12)
    rsi = 100.0 - 100.0 / (1.0 + rs)
    df["rsi"] = rsi

    # RSI divergence proxy: price makes new low/high but RSI does not
    df["rsi_5_min"]   = rsi.rolling(5).min()
    df["close_5_min"] = close.rolling(5).min()
    df["rsi_5_max"]   = rsi.rolling(5).max()
    df["close_5_max"] = close.rolling(5).max()

    # ── MACD ─────────────────────────────────────────────────────────────────
    ema_fast   = close.ewm(span=12, adjust=False).mean()
    ema_slow   = close.ewm(span=26, adjust=False).mean()
    macd_line  = ema_fast - ema_slow
    macd_signal = macd_line.ewm(span=9, adjust=False).mean()
    macd_hist  = macd_line - macd_signal

    df["macd_line"]   = macd_line
    df["macd_signal"] = macd_signal
    df["macd_hist"]   = macd_hist
    df["macd_hist_chg"] = macd_hist.diff()

    # ── Volume-like proxy: bar range ─────────────────────────────────────────
    bar_range = high - low
    df["bar_range"]      = bar_range
    df["bar_range_norm"] = bar_range / (atr + 1e-12)

    # ── Candle body direction & size ─────────────────────────────────────────
    body = close - open_
    df["body"]       = body
    df["body_norm"]  = body / (atr + 1e-12)
    df["body_dir"]   = np.where(body > 0, 1.0, np.where(body < 0, -1.0, 0.0))

    # ── Upper / lower wick ───────────────────────────────────────────────────
    df["upper_wick"] = high - pd.concat([close, open_], axis=1).max(axis=1)
    df["lower_wick"] = pd.concat([close, open_], axis=1).min(axis=1) - low

    # ── SMA trend context ─────────────────────────────────────────────────────
    sma_50  = close.rolling(50).mean()
    sma_200 = close.rolling(200).mean()
    df["sma_50"]         = sma_50
    df["sma_200"]        = sma_200
    df["close_vs_sma50"] = (close - sma_50) / (sma_50 + 1e-12)
    df["sma50_vs_sma200"] = (sma_50 - sma_200) / (sma_200 + 1e-12)

    # ── Volatility regime ────────────────────────────────────────────────────
    natr_ma = natr.rolling(50).mean()
    df["natr_ratio"] = natr / (natr_ma + 1e-12)   # >1 = above-avg vol

    # ── Mean-reversion distance ───────────────────────────────────────────────
    df["z_score_20"] = (close - bb_mid) / (bb_sigma + 1e-12)

    # ── Rolling realized vol ─────────────────────────────────────────────────
    df["rvol_10"] = df["ret_1"].rolling(10).std()
    df["rvol_20"] = df["ret_1"].rolling(20).std()

    # ── ATR-normalised returns ────────────────────────────────────────────────
    df["ret_1_natr"] = df["ret_1"] / (natr + 1e-12)

    # ── Lagged features ───────────────────────────────────────────────────────
    for lag in [1, 2, 3, 5]:
        df[f"bb_pct_lag{lag}"]    = bb_pct.shift(lag)
        df[f"bb_width_lag{lag}"]  = bb_width.shift(lag)
        df[f"rsi_lag{lag}"]       = rsi.shift(lag)
        df[f"macd_hist_lag{lag}"] = macd_hist.shift(lag)

    # ── Hour-of-day (cyclical) ────────────────────────────────────────────────
    hour = pd.Series(df.index.hour, index=df.index).astype(float)
    df["hour_sin"] = np.sin(2 * np.pi * hour / 24.0)
    df["hour_cos"] = np.cos(2 * np.pi * hour / 24.0)

    # ── Day-of-week (cyclical) ────────────────────────────────────────────────
    dow = pd.Series(df.index.dayofweek, index=df.index).astype(float)
    df["dow_sin"] = np.sin(2 * np.pi * dow / 5.0)
    df["dow_cos"] = np.cos(2 * np.pi * dow / 5.0)

    # ── Fill NaN from warm-up ────────────────────────────────────────────────
    df = df.bfill().ffill()

    return df


# SECTION 2 — STRATEGY CONFIG
def strategy_config():
    return {
        "title": "GBP/USD BB Squeeze Breakout (GradientBoosting)",

        "model_type": "GradientBoostingClassifier",
        "model_params": {
            "n_estimators":      400,
            "max_depth":         4,
            "learning_rate":     0.04,
            "subsample":         0.75,
            "min_samples_leaf":  20,
            "min_samples_split": 40,
            "max_features":      "sqrt",
            "n_iter_no_change":  30,
            "validation_fraction": 0.1,
            "tol":               1e-4,
            "random_state":      42,
        },

        "signal_threshold": 0.55,
        "direction":        "both",
        "stop_loss":        0.005,
        "take_profit":      0.010,
        "cooldown":         0,
        "max_positions":    1,
        "on_opposite":      "reverse",
        "session_filter":   [6, 20],
        "min_atr":          0.0002,
        "trend_filter":     None,
        "target_horizon":   4,

        "objective": (
            "Maximize risk-adjusted return (Sharpe / Calmar). "
            "GradientBoostingClassifier chosen for its strong performance on "
            "tabular financial data with noisy labels. Shallow trees (max_depth=4) "
            "with shrinkage (lr=0.04) and subsample=0.75 reduce overfitting. "
            "Early stopping (n_iter_no_change=30) prevents over-training. "
            "SL=0.5%, TP=1.0% gives a 1:2 risk/reward ratio. "
            "Session filter 06-20 UTC captures London + New York overlap for GBP/USD."
        ),
        "notes": (
            "Core signal: BB squeeze (narrow band width) followed by expansion "
            "breakout, confirmed by MACD histogram direction and RSI. "
            "ATR filter ensures minimum volatility for entries. "
            "Lagged BB features capture the squeeze build-up dynamic. "
            "Z-score and normalized returns give the model mean-reversion context. "
            "Cyclical time features allow the model to learn intraday seasonality."
        ),
    }

# ╔══════════════════════════════════════════════════════════════╗
# ║                  STRATEGY REQUEST LOG                       ║
# ╚══════════════════════════════════════════════════════════════╝
# Generated        : 2026-05-06 01:27:43
# Model            : XGBoost
# Feature Eng.     : SMA (20,50,200) + Auto-add features: ON
# Signal / Entry   : Enter when model confidence > threshold; exit on opposite signal or SL/TP
# Optimization     : Maximize risk-adjusted return
# Risk Mgmt        : Stop loss 0.5%, Take profit 1.0%
# Risk Filter      : —
# ══════════════════════════════════════════════════════════════

# ============================================================
# SECTION 0 — IMPORTS & CONSTANTS
import numpy as np
import pandas as pd

DATA_PATH = "/root/Desktop/QuantifyMe/data/ohlc/GBPUSD_15min.parquet"
START_DATE = "2025-04-24"
END_DATE   = "2026-04-24"
VALIDATION_DATE = ""
TRAIN_SPLIT = 0.7

# SECTION 1 — FEATURE ENGINEERING
def feature_engineering(df, close, open_, high, low):
    # ── Required SMAs and distance metrics ──────────────────────────────────
    for p in [20, 50, 200]:
        sma = close.rolling(p).mean()
        df[f"sma_{p}"] = sma
        df[f"dm_sma_{p}"] = (close - sma) / sma

    # ── SMA slope (momentum of the moving average itself) ───────────────────
    for p in [20, 50]:
        sma = close.rolling(p).mean()
        df[f"sma_{p}_slope"] = sma.diff(5) / sma.shift(5)

    # ── SMA cross signals ────────────────────────────────────────────────────
    sma20 = close.rolling(20).mean()
    sma50 = close.rolling(50).mean()
    sma200 = close.rolling(200).mean()
    df["sma20_50_cross"] = (sma20 - sma50) / sma50
    df["sma50_200_cross"] = (sma50 - sma200) / sma200
    df["sma20_200_cross"] = (sma20 - sma200) / sma200

    # ── Price momentum over multiple horizons ────────────────────────────────
    for lag in [1, 3, 6, 12, 24, 48]:
        df[f"ret_{lag}"] = close.pct_change(lag)

    # ── Volatility: rolling standard deviation of returns ───────────────────
    ret1 = close.pct_change(1)
    for w in [10, 20, 40]:
        df[f"vol_{w}"] = ret1.rolling(w).std()

    # ── ATR (Average True Range, normalised) ─────────────────────────────────
    tr = pd.concat([
        high - low,
        (high - close.shift(1)).abs(),
        (low  - close.shift(1)).abs()
    ], axis=1).max(axis=1)
    for w in [14, 28]:
        atr = tr.ewm(span=w, adjust=False).mean()
        df[f"natr_{w}"] = atr / close

    # ── Bollinger Bands (20-period, 2σ) ──────────────────────────────────────
    bb_mid  = close.rolling(20).mean()
    bb_std  = close.rolling(20).std()
    bb_up   = bb_mid + 2 * bb_std
    bb_lo   = bb_mid - 2 * bb_std
    bb_width = (bb_up - bb_lo) / bb_mid
    df["bb_pct_b"]  = (close - bb_lo) / (bb_up - bb_lo + 1e-12)
    df["bb_width"]  = bb_width
    df["bb_squeeze"]= np.where(bb_width < bb_width.rolling(50).mean(), 1.0, 0.0)

    # ── Keltner Channel (for squeeze confirmation) ───────────────────────────
    kc_mid  = close.ewm(span=20, adjust=False).mean()
    kc_atr  = tr.ewm(span=20, adjust=False).mean()
    kc_up   = kc_mid + 1.5 * kc_atr
    kc_lo   = kc_mid - 1.5 * kc_atr
    df["kc_pct"] = (close - kc_lo) / (kc_up - kc_lo + 1e-12)

    # ── RSI (Wilder) ─────────────────────────────────────────────────────────
    def wilder_rsi(src, period):
        delta = src.diff(1)
        gain  = delta.clip(lower=0)
        loss  = (-delta).clip(lower=0)
        avg_g = gain.ewm(alpha=1/period, adjust=False).mean()
        avg_l = loss.ewm(alpha=1/period, adjust=False).mean()
        rs    = avg_g / (avg_l + 1e-12)
        return 100 - 100 / (1 + rs)

    rsi14 = wilder_rsi(close, 14)
    rsi6  = wilder_rsi(close, 6)
    rsi28 = wilder_rsi(close, 28)
    df["rsi14"] = rsi14 / 100.0
    df["rsi6"]  = rsi6  / 100.0
    df["rsi28"] = rsi28 / 100.0
    df["rsi14_slope"] = rsi14.diff(3) / 100.0

    # RSI divergence proxy: price new high/low but RSI doesn't confirm
    price_high_12 = close.rolling(12).max()
    price_low_12  = close.rolling(12).min()
    rsi_high_12   = rsi14.rolling(12).max()
    rsi_low_12    = rsi14.rolling(12).min()
    df["rsi_bear_div"] = np.where(
        (close >= price_high_12 * 0.999) & (rsi14 < rsi_high_12 * 0.97), 1.0, 0.0)
    df["rsi_bull_div"] = np.where(
        (close <= price_low_12  * 1.001) & (rsi14 > rsi_low_12  * 1.03), 1.0, 0.0)

    # ── MACD ─────────────────────────────────────────────────────────────────
    ema12  = close.ewm(span=12, adjust=False).mean()
    ema26  = close.ewm(span=26, adjust=False).mean()
    macd   = ema12 - ema26
    signal = macd.ewm(span=9, adjust=False).mean()
    hist   = macd - signal
    df["macd_norm"]    = macd   / close
    df["macd_sig_norm"]= signal / close
    df["macd_hist_norm"]= hist  / close
    df["macd_hist_slope"] = hist.diff(2) / close

    # ── Stochastic Oscillator ─────────────────────────────────────────────────
    for k_period in [14, 5]:
        lo_k = low.rolling(k_period).min()
        hi_k = high.rolling(k_period).max()
        stoch_k = (close - lo_k) / (hi_k - lo_k + 1e-12) * 100
        stoch_d = stoch_k.rolling(3).mean()
        df[f"stoch_k_{k_period}"] = stoch_k / 100.0
        df[f"stoch_d_{k_period}"] = stoch_d / 100.0
        df[f"stoch_kd_{k_period}"] = (stoch_k - stoch_d) / 100.0

    # ── Williams %R ───────────────────────────────────────────────────────────
    hi14 = high.rolling(14).max()
    lo14 = low.rolling(14).min()
    df["williams_r"] = (hi14 - close) / (hi14 - lo14 + 1e-12)

    # ── CCI (Commodity Channel Index) ────────────────────────────────────────
    tp = (high + low + close) / 3.0
    tp_sma = tp.rolling(20).mean()
    tp_mad  = tp.rolling(20).apply(lambda x: np.mean(np.abs(x - x.mean())), raw=True)
    df["cci"] = (tp - tp_sma) / (0.015 * tp_mad + 1e-12) / 100.0

    # ── Volume-like proxy: candle body and wick ratios ────────────────────────
    candle_range = (high - low).replace(0, np.nan)
    df["body_ratio"]  = (close - open_).abs() / candle_range
    df["upper_wick"]  = (high - pd.concat([close, open_], axis=1).max(axis=1)) / candle_range
    df["lower_wick"]  = (pd.concat([close, open_], axis=1).min(axis=1) - low) / candle_range
    df["bull_candle"]  = np.where(close > open_, 1.0, 0.0)

    # ── Mean reversion signal: z-score of close vs SMA20 ────────────────────
    df["zscore_20"] = (close - sma20) / (close.rolling(20).std() + 1e-12)
    df["zscore_50"] = (close - sma50) / (close.rolling(50).std() + 1e-12)

    # ── Trend strength: ADX proxy ─────────────────────────────────────────────
    plus_dm  = (high.diff(1)).clip(lower=0)
    minus_dm = (-low.diff(1)).clip(lower=0)
    overlap  = pd.concat([plus_dm, minus_dm], axis=1).min(axis=1)
    plus_dm  = plus_dm - overlap
    minus_dm = minus_dm - overlap
    atr14 = tr.ewm(span=14, adjust=False).mean()
    plus_di  = 100 * plus_dm.ewm(span=14, adjust=False).mean()  / (atr14 + 1e-12)
    minus_di = 100 * minus_dm.ewm(span=14, adjust=False).mean() / (atr14 + 1e-12)
    dx = (plus_di - minus_di).abs() / (plus_di + minus_di + 1e-12) * 100
    adx = dx.ewm(span=14, adjust=False).mean()
    df["adx"]      = adx / 100.0
    df["plus_di"]  = plus_di / 100.0
    df["minus_di"] = minus_di / 100.0
    df["di_diff"]  = (plus_di - minus_di) / 100.0

    # ── Regime detection: above/below long-term SMA ──────────────────────────
    df["bull_regime"]  = np.where(close > sma200, 1.0, 0.0)
    df["mid_regime"]   = np.where(close > sma50,  1.0, 0.0)

    # ── Lag features (auto-regressive) ───────────────────────────────────────
    for col, lags in [("rsi14", [1, 2, 4]), ("macd_hist_norm", [1, 2]), ("bb_pct_b", [1, 2])]:
        for lag in lags:
            df[f"{col}_lag{lag}"] = df[col].shift(lag)

    # ── Time-of-day features ─────────────────────────────────────────────────
    if hasattr(df.index, "hour"):
        df["hour_sin"] = np.sin(2 * np.pi * df.index.hour / 24.0)
        df["hour_cos"] = np.cos(2 * np.pi * df.index.hour / 24.0)
        df["dow_sin"]  = np.sin(2 * np.pi * df.index.dayofweek / 5.0)
        df["dow_cos"]  = np.cos(2 * np.pi * df.index.dayofweek / 5.0)

    # ── Fill NaN from warm-up ────────────────────────────────────────────────
    df = df.bfill().ffill()
    return df


# SECTION 2 — STRATEGY CONFIG
def strategy_config():
    return {
        "title":            "GBP/USD SMA Trend + Multi-Indicator XGBoost Classifier",
        "model_type":       "XGBClassifier",
        "model_params": {
            "n_estimators":     600,
            "max_depth":        4,
            "learning_rate":    0.03,
            "subsample":        0.75,
            "colsample_bytree": 0.70,
            "min_child_weight": 5,
            "gamma":            0.10,
            "reg_alpha":        0.10,
            "reg_lambda":       1.50,
            "objective":        "binary:logistic",
            "tree_method":      "hist",
            "random_state":     42,
            "n_jobs":           -1,
        },
        "signal_threshold": 0.56,
        "direction":        "both",
        "stop_loss":        0.005,
        "take_profit":      0.010,
        "cooldown":         0,
        "max_positions":    1,
        "on_opposite":      "reverse",
        "session_filter":   [6, 18],
        "min_atr":          0.0003,
        "trend_filter":     "sma_50",
        "target_horizon":   4,
        "objective": (
            "Maximize risk-adjusted return on GBP/USD 15-min bars. "
            "Strategy combines required SMA (20/50/200) distance and cross features "
            "with ADX trend strength, RSI divergence, Bollinger squeeze, Keltner, "
            "MACD histogram slope, Stochastic, CCI, Williams %R, and candle-structure "
            "ratios. XGBoost with strong regularisation and subsampling prevents "
            "overfitting on the relatively short 1-year window. "
            "Session filter 06-18 UTC keeps execution in liquid London/NY hours; "
            "0.5% SL and 1.0% TP yield 1:2 R:R; sma_50 trend filter aligns trades "
            "with intermediate momentum to improve win rate and Sharpe."
        ),
        "notes": (
            "Differs from prior RSI/MACD/BB/Stoch attempts by: (1) foregrounding "
            "SMA cross and distance features as primary trend signals; (2) adding "
            "ADX-based regime and DI differential; (3) including RSI divergence "
            "proxy flags; (4) z-score mean-reversion features; (5) candle body/wick "
            "structure ratios as micro-structure proxies; (6) time-of-day cyclical "
            "encoding; (7) heavier regularisation (gamma, alpha, lambda) and higher "
            "min_child_weight to reduce variance on the thin dataset."
        ),
    }

# ╔══════════════════════════════════════════════════════════════╗
# ║                  STRATEGY REQUEST LOG                       ║
# ╚══════════════════════════════════════════════════════════════╝
# Generated        : 2026-05-06 01:47:56
# Model            : Gradient Boosting
# Feature Eng.     : SMA (20,50,200) + Auto-add features: ON
# Signal / Entry   : Enter when model confidence > threshold; exit on opposite signal or SL/TP
# Optimization     : Maximize risk-adjusted return
# Risk Mgmt        : Stop loss 0.5%, Take profit 1.0%
# Risk Filter      : —
# ══════════════════════════════════════════════════════════════

# ============================================================
# SECTION 0 — IMPORTS & CONSTANTS
import numpy as np
import pandas as pd

DATA_PATH = "/root/Desktop/QuantifyMe/data/ohlc/GBPUSD_15min.parquet"
START_DATE = "2025-04-24"
END_DATE   = "2026-04-24"
VALIDATION_DATE = ""
TRAIN_SPLIT = 0.7

# SECTION 1 — FEATURE ENGINEERING
def feature_engineering(df, close, open_, high, low):
    # ── SMA features (required) ──────────────────────────────────────────────
    for period in [20, 50, 200]:
        sma = close.rolling(period).mean()
        df[f"sma_{period}"] = sma
        df[f"dm_sma_{period}"] = (close - sma) / sma

    # ── SMA crossover signals ────────────────────────────────────────────────
    sma_20 = close.rolling(20).mean()
    sma_50 = close.rolling(50).mean()
    sma_200 = close.rolling(200).mean()

    df["sma_20_50_cross"] = np.where(sma_20 > sma_50, 1.0, -1.0)
    df["sma_20_200_cross"] = np.where(sma_20 > sma_200, 1.0, -1.0)
    df["sma_50_200_cross"] = np.where(sma_50 > sma_200, 1.0, -1.0)

    # ── Price momentum features ──────────────────────────────────────────────
    for lag in [1, 2, 4, 8, 16]:
        df[f"ret_{lag}"] = close.pct_change(lag)

    # ── Volatility: rolling std of returns ──────────────────────────────────
    ret_1 = close.pct_change(1)
    for window in [8, 20, 50]:
        df[f"vol_{window}"] = ret_1.rolling(window).std()

    # ── ATR (Average True Range) ─────────────────────────────────────────────
    tr = pd.concat([
        high - low,
        (high - close.shift(1)).abs(),
        (low  - close.shift(1)).abs()
    ], axis=1).max(axis=1)

    for atr_period in [14, 50]:
        atr = tr.rolling(atr_period).mean()
        df[f"atr_{atr_period}"] = atr
        df[f"natr_{atr_period}"] = atr / close

    # ── RSI ──────────────────────────────────────────────────────────────────
    for rsi_period in [14, 28]:
        delta = close.diff()
        gain  = delta.clip(lower=0).rolling(rsi_period).mean()
        loss  = (-delta.clip(upper=0)).rolling(rsi_period).mean()
        rs    = gain / (loss + 1e-10)
        df[f"rsi_{rsi_period}"] = 100 - (100 / (1 + rs))

    # ── MACD ─────────────────────────────────────────────────────────────────
    ema_12 = close.ewm(span=12, adjust=False).mean()
    ema_26 = close.ewm(span=26, adjust=False).mean()
    macd_line   = ema_12 - ema_26
    signal_line = macd_line.ewm(span=9, adjust=False).mean()
    df["macd"]           = macd_line
    df["macd_signal"]    = signal_line
    df["macd_hist"]      = macd_line - signal_line
    df["macd_hist_norm"] = (macd_line - signal_line) / (close + 1e-10)

    # ── Bollinger Bands ───────────────────────────────────────────────────────
    for bb_period in [20, 50]:
        bb_mid  = close.rolling(bb_period).mean()
        bb_std  = close.rolling(bb_period).std()
        bb_upper = bb_mid + 2.0 * bb_std
        bb_lower = bb_mid - 2.0 * bb_std
        bb_width = (bb_upper - bb_lower) / (bb_mid + 1e-10)
        bb_pos   = (close - bb_lower) / (bb_upper - bb_lower + 1e-10)
        df[f"bb_width_{bb_period}"] = bb_width
        df[f"bb_pos_{bb_period}"]   = bb_pos

    # ── Stochastic Oscillator ────────────────────────────────────────────────
    for stoch_period in [14, 28]:
        lowest_low   = low.rolling(stoch_period).min()
        highest_high = high.rolling(stoch_period).max()
        stoch_k = (close - lowest_low) / (highest_high - lowest_low + 1e-10) * 100
        stoch_d = stoch_k.rolling(3).mean()
        df[f"stoch_k_{stoch_period}"] = stoch_k
        df[f"stoch_d_{stoch_period}"] = stoch_d

    # ── Rate of Change (ROC) ──────────────────────────────────────────────────
    for roc_period in [5, 10, 20]:
        df[f"roc_{roc_period}"] = close.pct_change(roc_period)

    # ── Candle body and shadow features ──────────────────────────────────────
    body        = (close - open_).abs()
    candle_range = (high - low).abs()
    df["body_ratio"]       = body / (candle_range + 1e-10)
    df["upper_shadow"]     = (high - pd.concat([close, open_], axis=1).max(axis=1)) / (candle_range + 1e-10)
    df["lower_shadow"]     = (pd.concat([close, open_], axis=1).min(axis=1) - low) / (candle_range + 1e-10)
    df["bullish_candle"]   = np.where(close > open_, 1.0, -1.0)

    # ── Volume-proxy: candle range as volatility proxy ────────────────────────
    df["range_norm"] = candle_range / (close + 1e-10)
    df["range_ma_ratio"] = candle_range / (candle_range.rolling(20).mean() + 1e-10)

    # ── Lag features for return predictors ───────────────────────────────────
    for col_lag in ["rsi_14", "macd_hist", "bb_pos_20"]:
        for lag in [1, 2, 3]:
            df[f"{col_lag}_lag{lag}"] = df[col_lag].shift(lag)

    # ── Distance of close from recent high/low ────────────────────────────────
    for lookback in [10, 20, 50]:
        roll_high = high.rolling(lookback).max()
        roll_low  = low.rolling(lookback).min()
        df[f"dist_high_{lookback}"] = (close - roll_high) / (roll_high + 1e-10)
        df[f"dist_low_{lookback}"]  = (close - roll_low)  / (roll_low  + 1e-10)

    # ── Trend strength: ADX proxy ─────────────────────────────────────────────
    adx_period = 14
    tr_adx  = tr.copy()
    plus_dm  = pd.Series(np.where((high.diff() > 0) & (high.diff() > -low.diff()), high.diff(), 0.0), index=close.index)
    minus_dm = pd.Series(np.where((-low.diff() > 0) & (-low.diff() > high.diff()), -low.diff(), 0.0), index=close.index)
    atr_adx  = tr_adx.rolling(adx_period).mean()
    plus_di  = 100 * plus_dm.rolling(adx_period).mean()  / (atr_adx + 1e-10)
    minus_di = 100 * minus_dm.rolling(adx_period).mean() / (atr_adx + 1e-10)
    dx       = (100 * (plus_di - minus_di).abs() / (plus_di + minus_di + 1e-10))
    df["adx"]       = dx.rolling(adx_period).mean()
    df["plus_di"]   = plus_di
    df["minus_di"]  = minus_di

    # ── Fill NaN from indicator warm-up ──────────────────────────────────────
    df = df.bfill().ffill()

    return df


# SECTION 2 — STRATEGY CONFIG
def strategy_config():
    return {
        "title":            "GBP/USD SMA Trend Gradient Boosting Risk-Adj",
        "model_type":       "GradientBoostingClassifier",
        "model_params":     {
            "n_estimators":     400,
            "max_depth":        4,
            "learning_rate":    0.04,
            "subsample":        0.75,
            "min_samples_leaf": 20,
            "max_features":     "sqrt",
            "n_iter_no_change": 30,
            "validation_fraction": 0.1,
            "tol":              1e-4,
            "random_state":     42,
        },
        "signal_threshold": 0.57,
        "direction":        "both",
        "stop_loss":        0.005,
        "take_profit":      0.010,
        "cooldown":         0,
        "max_positions":    1,
        "on_opposite":      "reverse",
        "session_filter":   [6, 18],
        "min_atr":          0.0002,
        "trend_filter":     "sma_50",
        "target_horizon":   4,
        "objective": (
            "Maximize risk-adjusted return (Sharpe/Calmar) on GBP/USD 15-min data. "
            "GradientBoostingClassifier chosen for its strong bias-variance tradeoff "
            "on medium-sized tabular datasets without needing GPU. "
            "Hyperparameters: moderate depth=4 prevents overfitting, learning_rate=0.04 "
            "with 400 estimators balances convergence vs generalisation, subsample=0.75 "
            "adds stochasticity to reduce variance, min_samples_leaf=20 enforces statistical "
            "significance at each leaf. Early stopping via n_iter_no_change guards against "
            "overfit on the training fold. Signal threshold 0.57 filters marginal signals "
            "to improve precision. SL=0.5%, TP=1.0% gives 1:2 RR. Session filter 6-18 UTC "
            "covers London+NY overlap — highest GBP/USD liquidity and tighter spreads. "
            "sma_50 trend filter ensures we only trade in the direction of medium-term trend, "
            "reducing whipsaw losses. target_horizon=4 bars (1 hour) gives the model enough "
            "time for moves to develop while staying relevant for intraday trading."
        ),
        "notes": (
            "Features: SMA 20/50/200 with distance metrics (core requirement), RSI 14/28, "
            "MACD, Bollinger Bands 20/50, Stochastic 14/28, ATR 14/50, NATR, ROC, ADX, "
            "candle body/shadow ratios, lagged RSI/MACD/BB features, distance from rolling "
            "high/low, SMA crossover signals, multi-lag return features. "
            "All features are backward-looking only (no lookahead bias). "
            "on_opposite=reverse for fast trend-following entries without missing reversals."
        ),
    }

# ╔══════════════════════════════════════════════════════════════╗
# ║                  STRATEGY REQUEST LOG                       ║
# ╚══════════════════════════════════════════════════════════════╝
# Generated        : 2026-05-06 01:23:23
# Model            : XGBoost
# Feature Eng.     : RSI 14, MACD (12,26,9) + Auto-add features: ON
# Signal / Entry   : Enter when model confidence > threshold; exit on opposite signal or SL/TP
# Optimization     : Maximize risk-adjusted return
# Risk Mgmt        : Stop loss 0.5%, Take profit 1.0%
# Risk Filter      : —
# ══════════════════════════════════════════════════════════════

# ============================================================
# SECTION 0 — IMPORTS & CONSTANTS
import numpy as np
import pandas as pd

DATA_PATH = "/root/Desktop/QuantifyMe/data/ohlc/GBPUSD_15min.parquet"
START_DATE = "2025-04-24"
END_DATE   = "2026-04-24"
VALIDATION_DATE = ""
TRAIN_SPLIT = 0.7

# SECTION 1 — FEATURE ENGINEERING
def feature_engineering(df, close, open_, high, low):

    # --- RSI 14 ---
    delta = close.diff()
    gain = delta.clip(lower=0)
    loss = -delta.clip(upper=0)
    avg_gain = gain.ewm(com=13, min_periods=14).mean()
    avg_loss = loss.ewm(com=13, min_periods=14).mean()
    rs = avg_gain / (avg_loss + 1e-12)
    df["rsi_14"] = 100 - (100 / (1 + rs))

    # RSI derived features
    df["rsi_zscore"] = (df["rsi_14"] - df["rsi_14"].rolling(50).mean()) / (df["rsi_14"].rolling(50).std() + 1e-12)
    df["rsi_slope"] = df["rsi_14"].diff(3)
    df["rsi_above_50"] = np.where(df["rsi_14"] > 50, 1, 0)
    df["rsi_overbought"] = np.where(df["rsi_14"] > 70, 1, 0)
    df["rsi_oversold"] = np.where(df["rsi_14"] < 30, 1, 0)
    # RSI divergence proxy: price direction vs RSI direction
    price_dir_3 = np.sign(close.diff(3))
    rsi_dir_3 = np.sign(df["rsi_14"].diff(3))
    df["rsi_divergence"] = np.where(price_dir_3 != rsi_dir_3, 1, 0)

    # --- MACD (12, 26, 9) ---
    ema12 = close.ewm(span=12, adjust=False).mean()
    ema26 = close.ewm(span=26, adjust=False).mean()
    macd_line = ema12 - ema26
    signal_line = macd_line.ewm(span=9, adjust=False).mean()
    macd_hist = macd_line - signal_line
    df["macd_line"] = macd_line
    df["macd_signal"] = signal_line
    df["macd_hist"] = macd_hist

    # MACD derived features
    df["macd_hist_slope"] = macd_hist.diff(2)
    df["macd_cross_up"] = np.where((macd_line > signal_line) & (macd_line.shift(1) <= signal_line.shift(1)), 1, 0)
    df["macd_cross_dn"] = np.where((macd_line < signal_line) & (macd_line.shift(1) >= signal_line.shift(1)), 1, 0)
    df["macd_hist_positive"] = np.where(macd_hist > 0, 1, 0)
    df["macd_hist_expanding"] = np.where(macd_hist.abs() > macd_hist.abs().shift(1), 1, 0)
    df["macd_normalized"] = macd_line / (close + 1e-12)

    # --- ATR 14 ---
    tr1 = high - low
    tr2 = (high - close.shift(1)).abs()
    tr3 = (low - close.shift(1)).abs()
    tr = pd.concat([tr1, tr2, tr3], axis=1).max(axis=1)
    atr14 = tr.ewm(com=13, min_periods=14).mean()
    df["atr_14"] = atr14
    df["natr_14"] = atr14 / (close + 1e-12)

    # ATR regime: high vs low volatility
    atr_ma = atr14.rolling(50).mean()
    df["atr_high_vol"] = np.where(atr14 > atr_ma * 1.2, 1, 0)
    df["atr_low_vol"] = np.where(atr14 < atr_ma * 0.8, 1, 0)

    # --- Bollinger Bands (20, 2) ---
    bb_mid = close.rolling(20).mean()
    bb_std = close.rolling(20).std()
    bb_upper = bb_mid + 2 * bb_std
    bb_lower = bb_mid - 2 * bb_std
    df["bb_pct_b"] = (close - bb_lower) / (bb_upper - bb_lower + 1e-12)
    df["bb_width"] = (bb_upper - bb_lower) / (bb_mid + 1e-12)
    df["bb_squeeze"] = np.where(df["bb_width"] < df["bb_width"].rolling(50).quantile(0.2), 1, 0)
    df["bb_upper_touch"] = np.where(close >= bb_upper * 0.999, 1, 0)
    df["bb_lower_touch"] = np.where(close <= bb_lower * 1.001, 1, 0)

    # --- Keltner Channel (20, 1.5x ATR) ---
    kc_mid = close.ewm(span=20, adjust=False).mean()
    kc_upper = kc_mid + 1.5 * atr14
    kc_lower = kc_mid - 1.5 * atr14
    df["kc_pct"] = (close - kc_lower) / (kc_upper - kc_lower + 1e-12)
    # Squeeze: BB inside KC
    df["kc_bb_squeeze"] = np.where((bb_upper < kc_upper) & (bb_lower > kc_lower), 1, 0)

    # --- Volume-like proxy: bar range & body ---
    bar_range = high - low
    bar_body = (close - open_).abs()
    df["range_norm"] = bar_range / (atr14 + 1e-12)
    df["body_ratio"] = bar_body / (bar_range + 1e-12)
    df["close_position"] = (close - low) / (bar_range + 1e-12)
    df["bullish_bar"] = np.where(close > open_, 1, 0)

    # --- Momentum & ROC ---
    df["roc_5"] = close.pct_change(5)
    df["roc_10"] = close.pct_change(10)
    df["roc_20"] = close.pct_change(20)
    df["momentum_10"] = close - close.shift(10)
    df["momentum_20"] = close - close.shift(20)

    # --- Moving Average features ---
    ema8 = close.ewm(span=8, adjust=False).mean()
    ema21 = close.ewm(span=21, adjust=False).mean()
    ema50 = close.ewm(span=50, adjust=False).mean()
    sma20 = close.rolling(20).mean()
    sma50 = close.rolling(50).mean()
    sma100 = close.rolling(100).mean()

    df["ema8_21_gap"] = (ema8 - ema21) / (close + 1e-12)
    df["ema21_50_gap"] = (ema21 - ema50) / (close + 1e-12)
    df["price_vs_ema50"] = (close - ema50) / (close + 1e-12)
    df["price_vs_sma20"] = (close - sma20) / (close + 1e-12)
    df["price_vs_sma100"] = (close - sma100) / (close + 1e-12)
    df["ema8_slope"] = ema8.diff(3) / (close + 1e-12)
    df["ema21_slope"] = ema21.diff(3) / (close + 1e-12)
    df["ema8_above_ema21"] = np.where(ema8 > ema21, 1, 0)
    df["ema21_above_ema50"] = np.where(ema21 > ema50, 1, 0)
    df["triple_ma_align_bull"] = np.where((ema8 > ema21) & (ema21 > ema50), 1, 0)
    df["triple_ma_align_bear"] = np.where((ema8 < ema21) & (ema21 < ema50), 1, 0)

    # --- Stochastic %K %D (14, 3) ---
    lowest_low_14 = low.rolling(14).min()
    highest_high_14 = high.rolling(14).max()
    stoch_k = 100 * (close - lowest_low_14) / (highest_high_14 - lowest_low_14 + 1e-12)
    stoch_d = stoch_k.rolling(3).mean()
    df["stoch_k"] = stoch_k
    df["stoch_d"] = stoch_d
    df["stoch_kd_diff"] = stoch_k - stoch_d
    df["stoch_overbought"] = np.where(stoch_k > 80, 1, 0)
    df["stoch_oversold"] = np.where(stoch_k < 20, 1, 0)
    df["stoch_cross_up"] = np.where((stoch_k > stoch_d) & (stoch_k.shift(1) <= stoch_d.shift(1)), 1, 0)
    df["stoch_cross_dn"] = np.where((stoch_k < stoch_d) & (stoch_k.shift(1) >= stoch_d.shift(1)), 1, 0)

    # --- Williams %R (14) ---
    df["willr_14"] = -100 * (highest_high_14 - close) / (highest_high_14 - lowest_low_14 + 1e-12)

    # --- CCI (20) ---
    tp = (high + low + close) / 3
    tp_ma = tp.rolling(20).mean()
    tp_mad = tp.rolling(20).apply(lambda x: np.mean(np.abs(x - np.mean(x))), raw=True)
    df["cci_20"] = (tp - tp_ma) / (0.015 * tp_mad + 1e-12)
    df["cci_above_zero"] = np.where(df["cci_20"] > 0, 1, 0)
    df["cci_extreme_bull"] = np.where(df["cci_20"] > 100, 1, 0)
    df["cci_extreme_bear"] = np.where(df["cci_20"] < -100, 1, 0)

    # --- Donchian Channel (20) ---
    don_high = high.rolling(20).max()
    don_low = low.rolling(20).min()
    df["donchian_pct"] = (close - don_low) / (don_high - don_low + 1e-12)
    df["donchian_breakout_up"] = np.where(close >= high.rolling(20).max().shift(1), 1, 0)
    df["donchian_breakout_dn"] = np.where(close <= low.rolling(20).min().shift(1), 1, 0)

    # --- Price pattern features ---
    df["higher_high"] = np.where((high > high.shift(1)) & (high.shift(1) > high.shift(2)), 1, 0)
    df["lower_low"] = np.where((low < low.shift(1)) & (low.shift(1) < low.shift(2)), 1, 0)
    df["inside_bar"] = np.where((high < high.shift(1)) & (low > low.shift(1)), 1, 0)
    df["outside_bar"] = np.where((high > high.shift(1)) & (low < low.shift(1)), 1, 0)

    # --- Lag features for key indicators ---
    for lag in [1, 2, 3, 4]:
        df[f"rsi_14_lag{lag}"] = df["rsi_14"].shift(lag)
        df[f"macd_hist_lag{lag}"] = df["macd_hist"].shift(lag)
        df[f"bb_pct_b_lag{lag}"] = df["bb_pct_b"].shift(lag)

    # --- Interaction features (avoiding lookahead) ---
    df["rsi_macd_bull"] = np.where((df["rsi_14"] > 50) & (df["macd_hist"] > 0), 1, 0)
    df["rsi_macd_bear"] = np.where((df["rsi_14"] < 50) & (df["macd_hist"] < 0), 1, 0)
    df["rsi_bb_oversold_bounce"] = np.where((df["rsi_14"] < 35) & (df["bb_pct_b"] < 0.2), 1, 0)
    df["rsi_bb_overbought_fade"] = np.where((df["rsi_14"] > 65) & (df["bb_pct_b"] > 0.8), 1, 0)
    df["triple_bull"] = np.where(
        (df["rsi_14"] > 50) & (df["macd_hist"] > 0) & (df["stoch_k"] > 50), 1, 0
    )
    df["triple_bear"] = np.where(
        (df["rsi_14"] < 50) & (df["macd_hist"] < 0) & (df["stoch_k"] < 50), 1, 0
    )

    # --- Volatility regime ---
    realized_vol = close.pct_change().rolling(20).std() * np.sqrt(96)
    df["realized_vol_20"] = realized_vol
    df["vol_regime_high"] = np.where(realized_vol > realized_vol.rolling(100).median(), 1, 0)

    # --- Session-aware time features ---
    if hasattr(df.index, 'hour'):
        df["hour_sin"] = np.sin(2 * np.pi * df.index.hour / 24)
        df["hour_cos"] = np.cos(2 * np.pi * df.index.hour / 24)
        df["london_session"] = np.where((df.index.hour >= 7) & (df.index.hour < 16), 1, 0)
        df["ny_session"] = np.where((df.index.hour >= 13) & (df.index.hour < 21), 1, 0)
        df["overlap_session"] = np.where((df.index.hour >= 13) & (df.index.hour < 16), 1, 0)
        df["asian_session"] = np.where((df.index.hour >= 0) & (df.index.hour < 7), 1, 0)
        df["day_of_week"] = df.index.dayofweek
        df["dow_sin"] = np.sin(2 * np.pi * df["day_of_week"] / 5)
        df["dow_cos"] = np.cos(2 * np.pi * df["day_of_week"] / 5)

    df = df.bfill().ffill()
    return df

# SECTION 2 — STRATEGY CONFIG
def strategy_config():
    return {
        "title":            "GBP/USD RSI-MACD Momentum + Volatility Regime XGBoost",
        "model_type":       "XGBClassifier",
        "model_params":     {
            "n_estimators":     500,
            "max_depth":        4,
            "learning_rate":    0.03,
            "subsample":        0.75,
            "colsample_bytree": 0.65,
            "min_child_weight": 5,
            "gamma":            0.15,
            "reg_alpha":        0.3,
            "reg_lambda":       1.5,
            "scale_pos_weight": 1,
            "objective":        "binary:logistic",
            "tree_method":      "hist",
            "random_state":     42,
        },
        "signal_threshold": 0.56,
        "direction":        "both",
        "stop_loss":        0.005,
        "take_profit":      0.010,
        "cooldown":         0,
        "max_positions":    1,
        "on_opposite":      "reverse",
        "session_filter":   [7, 21],
        "min_atr":          0.0002,
        "trend_filter":     None,
        "target_horizon":   4,
        "objective":        (
            "Maximize risk-adjusted return (Sharpe/Calmar) by combining RSI momentum "
            "divergence, MACD histogram dynamics, Bollinger squeeze, Stochastic crossovers, "
            "volatility regime, and session-aware time features. XGBoost with moderate depth "
            "and strong regularization prevents overfitting on 15-min GBP/USD data. "
            "Signal threshold 0.56 filters weak signals, SL/TP at 0.5%/1.0% gives 1:2 RR."
        ),
        "notes":            (
            "Differentiating from prior attempts (PF=1.08) by: (1) adding Keltner Channel "
            "squeeze interaction with Bollinger, (2) CCI and Williams %R as confirmation, "
            "Donchian breakout detection, (3) session-aware features (London/NY/overlap), "
            "(4) richer MACD/RSI interaction flags, (5) realized volatility regime, "
            "(6) stronger XGBoost regularization (alpha=0.3, lambda=1.5, min_child=5) "
            "to reduce false signals in choppy GBP/USD conditions."
        ),
    }