BluefxPraise/The_Oracle

1"""Brute-force EMA / RSI / threshold parameter sweeps."""2 3import logging4from typing import Any, Dict, List5 6import numpy as np7import pandas as pd8 9import memory10from core.risk_engine import compute_sl_tp, atr_value11 12log = logging.getLogger("oracle.tuner")13 14 15def _ema(s: pd.Series, n: int) -> pd.Series:16    return s.ewm(span=n, adjust=False).mean()17 18 19def _rsi(s: pd.Series, n: int = 14) -> pd.Series:20    d = s.diff()21    g = d.where(d > 0, 0.0).rolling(n).mean()22    l = (-d.where(d < 0, 0.0)).rolling(n).mean()23    rs = g / l.replace(0, np.nan)24    return (100 - 100 / (1 + rs)).fillna(50)25 26 27def _simulate(pair: str, candles: List[Dict], ema_fast: int, ema_slow: int,28              rsi_n: int, threshold: float) -> Dict[str, Any]:29    if len(candles) < 80 or ema_fast >= ema_slow:30        return {"pnl": -1e9, "trades": 0, "winrate": 0}31    df = pd.DataFrame(candles)32    df["ef"] = _ema(df["c"], ema_fast)33    df["es"] = _ema(df["c"], ema_slow)34    df["rsi"] = _rsi(df["c"], rsi_n)35 36    trades = []37    open_t = None38    for i in range(60, len(candles) - 1):39        bar = candles[i + 1]40        if open_t:41            hi, lo = float(bar["h"]), float(bar["l"])42            t = open_t43            if t["dir"] == "BUY":44                if lo <= t["sl"]:45                    t["pnl"] = (t["sl"] - t["entry"]) / t["entry"]; t["w"] = 046                    trades.append(t); open_t = None; continue47                if hi >= t["tp"]:48                    t["pnl"] = (t["tp"] - t["entry"]) / t["entry"]; t["w"] = 149                    trades.append(t); open_t = None; continue50            else:51                if hi >= t["sl"]:52                    t["pnl"] = (t["entry"] - t["sl"]) / t["entry"]; t["w"] = 053                    trades.append(t); open_t = None; continue54                if lo <= t["tp"]:55                    t["pnl"] = (t["entry"] - t["tp"]) / t["entry"]; t["w"] = 156                    trades.append(t); open_t = None; continue57        if open_t is None:58            ef, es = df["ef"].iloc[i], df["es"].iloc[i]59            rsi = df["rsi"].iloc[i]60            close = float(candles[i]["c"])61            score = 0.062            direction = None63            if ef > es and rsi > 50:64                direction = "BUY"; score = (ef - es) / max(1e-9, close) * 1000 + (rsi - 50)65            elif ef < es and rsi < 50:66                direction = "SELL"; score = (es - ef) / max(1e-9, close) * 1000 + (50 - rsi)67            confidence = min(100.0, max(0.0, 50 + score))68            if direction and confidence >= threshold:69                atr = atr_value(candles[: i + 1], 14)70                sl, tp, _ = compute_sl_tp(direction, close, atr, pair)71                open_t = {"dir": direction, "entry": close, "sl": sl, "tp": tp}72 73    wins = sum(t["w"] for t in trades)74    pnl = sum(t["pnl"] for t in trades)75    return {76        "pnl": float(pnl), "trades": len(trades),77        "winrate": 100.0 * wins / max(1, len(trades)),78    }79 80 81def sweep(pair: str, candles: List[Dict[str, Any]]) -> Dict[str, Any]:82    ema_fasts = [8, 12, 16]83    ema_slows = [20, 26, 34]84    rsi_ns = [9, 14, 21]85    thresholds = [60, 70, 80, 90]86    best = {"pnl": -1e9, "params": None, "metrics": None}87    grid = []88    for ef in ema_fasts:89        for es in ema_slows:90            if es <= ef:91                continue92            for rn in rsi_ns:93                for th in thresholds:94                    m = _simulate(pair, candles, ef, es, rn, th)95                    grid.append({"ef": ef, "es": es, "rsi": rn, "thr": th, **m})96                    if m["pnl"] > best["pnl"] and m["trades"] >= 3:97                        best = {"pnl": m["pnl"], "params": {"ef": ef, "es": es, "rsi": rn, "thr": th},98                                "metrics": m}99    memory.log_calibration(100        kind="parameter_sweep",101        params={"pair": pair, "best": best["params"]},102        metrics={"best": best["metrics"], "grid_size": len(grid)},103        applied=False,104    )105    return {"best": best, "grid_size": len(grid)}106