Harwinder Sekhon
32ad73c59a
Owner decisions: research only (live scans stay on the CPU code), gpu_backtest.py stays the base reference, every research flag supported, the one exemption from instructions item 6 recorded in backtest/instructions_for_backtest.txt. - backtest/cuda_signal_kernels.py: the CUDA decision stage. K1 = engine.py + the GPUTechnicalEngine overrides (one thread per bar), K2 = analyze_from_dataframes end to end incl. gates 6b-6i, the four setup generators and the regime filter (one thread per primary bar), K3 = position_sizing.build_features, K4 cooldown, K5 prefix median, K6 recorder bar rows. Config packer with generated #defines; leaf module; --fmad=false; Python evaluation order kept (ternary min/max, pairwise sums, Neumaier sum(), exact round(x, 2)). - backtest/cudaBasedBacktesting.py: same CLI, cache files and writers as the reference; the host turns kernel codes into the exact strings; no CPU path. - test_all.py Module 37 (16 tests): every stage pinned against the live code; suite 552 / 552 on the GPU, without CUDA and with FB2_BACKEND=cpu. - Proof: five reference MD5s, four 6-month acceptance books (M5/M15, $0.65/$0.30), M15 OOS --export-features and --record --record-bars tables, ten research-flag variants, and the owner's five-year M5 and M15 books, all byte-identical to gpu_backtest.py. Wall time 0.8-6.6 s per book against 13-353 s. - CLAUDE.md sections 1, 3, 6, 7, 9, 10, 11 and the instructions addendum: any signal change is now a parity event for the twin.
1029 lines
52 KiB
Python
1029 lines
52 KiB
Python
#!/usr/bin/env python3
|
||
"""
|
||
cudaBasedBacktesting.py — the full-GPU twin of backtest/gpu_backtest.py.
|
||
|
||
Same command line, same cache files, same exported book (byte for byte —
|
||
the MD5 of the CSV is the acceptance gate), but every stage of the
|
||
computation runs on the GPU: the indicator precompute (shared with
|
||
gpu_backtest.py), the per-bar decision (backtest/cuda_signal_kernels.py, a
|
||
CUDA twin of engine.py + signal_aggregator.py + chart_patterns.py), the
|
||
entry cooldown, the decision-time features (--export-features), the
|
||
recorder rows (--record, --record-bars) and the trade resolution (shared).
|
||
|
||
Research tool only (owner decision 2026-09-18): live scans stay on the CPU
|
||
code and gpu_backtest.py stays in the tree as the base reference. Any change
|
||
to the signal logic is a parity event for this file — regenerate a book with
|
||
both scripts and show equal MD5s before using numbers from here as evidence
|
||
(backtest/instructions_for_backtest.txt).
|
||
|
||
What stays on the host, on purpose: loading and cleaning the parquet cache,
|
||
the calendar/session flags (pandas time-zone arithmetic over the whole
|
||
index), turning the kernels' numeric codes into the exact strings the
|
||
reference script writes (indicator descriptions, warnings, reasons), and
|
||
the report / CSV / parquet writers, all of which are reused from the
|
||
reference script unchanged.
|
||
|
||
There is no CPU fallback here by design: the twin IS the live code.
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import logging
|
||
import sys
|
||
import time
|
||
import types
|
||
from pathlib import Path
|
||
from typing import Dict, List, Optional, Sequence, Tuple
|
||
|
||
import numpy as np
|
||
import pandas as pd
|
||
|
||
PROJECT_ROOT = Path(__file__).resolve().parent.parent
|
||
if str(PROJECT_ROOT) not in sys.path:
|
||
sys.path.insert(0, str(PROJECT_ROOT))
|
||
|
||
from config import PAIR_INFO, PlatformConfig, Timeframe # noqa: E402
|
||
from providers import make_data_provider # noqa: E402
|
||
from providers.base import BaseDataProvider # noqa: E402
|
||
from signal_aggregator import SignalAggregator # noqa: E402 (fingerprints + the two static cleanups only)
|
||
import position_sizing # noqa: E402
|
||
from backtest import gpu_kernels as _gk # noqa: E402
|
||
from backtest import cuda_signal_kernels as ck # noqa: E402
|
||
from backtest import replay_record # noqa: E402
|
||
from backtest.gpu_backtest import ( # noqa: E402
|
||
BAR_SECONDS, _NY_TZ, BacktestConfig, Trade, _apply_config_overrides, _apply_fixed_rr_tp,
|
||
_cache_path, _DAY_NAMES, _history_pad_days, _parse_args, _reanchor_entries_to_raw, _resolve_pair,
|
||
_resolve_tf, compute_metrics, fetch_history, precompute_indicators, print_report,
|
||
print_trade_reasoning, simulate_trades, trades_to_dataframe,
|
||
)
|
||
|
||
logger = logging.getLogger("cudaBasedBacktesting")
|
||
BACKEND = _gk.BACKEND
|
||
|
||
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
# Text: the exact templates of engine.py / signal_aggregator.py
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
|
||
_RSI_SUFFIX = {
|
||
0: "Overbought", 1: "Oversold", 2: "Bullish momentum zone", 3: "Bearish momentum zone", 4: "Neutral",
|
||
5: "Pullback zone in uptrend (ideal entry)", 6: "Deep pullback in uptrend", 7: "Momentum ok in uptrend",
|
||
8: "Extended in uptrend, wait for pullback", 9: "Oversold in uptrend, possible reversal",
|
||
10: "Bounce zone in downtrend (ideal short)", 11: "Deep bounce in downtrend", 12: "Momentum ok in downtrend",
|
||
13: "Extended in downtrend, wait for bounce", 14: "Overbought in downtrend, possible reversal",
|
||
15: "Overbought while trend forming", 16: "Oversold while trend forming", 17: "No RSI edge while trend forms",
|
||
18: "Overbought in range", 19: "Oversold in range", 20: "Upper range zone", 21: "Lower range zone",
|
||
22: "Mid-range, no edge",
|
||
}
|
||
_MACD_DESC = {
|
||
0: "MACD histogram accelerating ↑", 1: "MACD histogram crossed above zero",
|
||
2: "MACD histogram accelerating ↓", 3: "MACD histogram crossed below zero",
|
||
4: "MACD histogram positive but fading", 5: "MACD histogram negative but fading",
|
||
6: "MACD histogram flat",
|
||
}
|
||
_BB_DESC = {0: "Price above upper BB", 1: "Price below lower BB", 2: "Price near upper BB",
|
||
3: "Price near lower BB", 4: "Price mid-band"}
|
||
_KC_DESC = {
|
||
0: "Keltner SQUEEZE RELEASED — bullish breakout", 1: "Keltner SQUEEZE RELEASED — bearish breakout",
|
||
2: "Keltner squeeze released — direction unclear", 3: "Keltner SQUEEZE active — breakout imminent",
|
||
4: "Price above Keltner upper — strong uptrend continuation",
|
||
5: "Price below Keltner lower — strong downtrend continuation",
|
||
6: "Price within Keltner Channels",
|
||
}
|
||
_EMA2_DESC = {
|
||
0: "EMA200 data insufficient", 1: "Full bullish: price > EMA50 > EMA200",
|
||
2: "Emerging uptrend: price > EMA50, EMA50 < EMA200", 3: "Full bearish: price < EMA50 < EMA200",
|
||
4: "Emerging downtrend: price < EMA50, EMA50 > EMA200", 5: "EMA alignment mixed",
|
||
}
|
||
_DIV_ITEMS = ("Bullish RSI divergence (reversal warning)", "Bearish RSI divergence (reversal warning)",
|
||
"Bullish MACD divergence (reversal warning)", "Bearish MACD divergence (reversal warning)")
|
||
_CS_ITEMS = ("Doji (indecision)", "Hammer (bullish reversal)", "Hanging Man (bearish warning)",
|
||
"Shooting Star (bearish reversal)", "Inverted Hammer (bullish)", "Bullish Engulfing",
|
||
"Bearish Engulfing", "Bullish Marubozu (strong buyers)", "Bearish Marubozu (strong sellers)",
|
||
"(at support)", "(at resistance)")
|
||
_ADX_STRENGTH = ("Very strong", "Strong", "Weak/Ranging")
|
||
_ADX_DIR = ("bullish", "bearish")
|
||
|
||
|
||
class K1Row:
|
||
"""One timeframe bar's K1 outputs as Python scalars (name -> value)."""
|
||
|
||
__slots__ = ("i", "f")
|
||
|
||
def __init__(self, ints: Dict[str, int], floats: Dict[str, float]):
|
||
self.i, self.f = ints, floats
|
||
|
||
|
||
def k1_rows(an: "ck.TFAnalysis", idx: np.ndarray) -> List[K1Row]:
|
||
"""Gather K1 outputs for the given bar indexes (one host transfer)."""
|
||
idx = np.asarray(idx, dtype=np.int64)
|
||
if len(idx) == 0:
|
||
return []
|
||
fi = an.OF[:, idx].get() # [K1F_N, m]
|
||
ii = an.OI[:, idx].get() # [K1I_N, m]
|
||
rows = []
|
||
for m in range(len(idx)):
|
||
rows.append(K1Row({n: int(ii[k, m]) for k, n in enumerate(ck.K1_INT_FIELDS)},
|
||
{n: float(fi[k, m]) for k, n in enumerate(ck.K1_FLOAT_FIELDS)}))
|
||
return rows
|
||
|
||
|
||
def describe_indicators(row: K1Row, ema_mode_2ema: bool) -> Dict[str, Dict]:
|
||
"""The scored indicators of one analysis in engine insertion order:
|
||
name -> {value, signal (int), confidence, description, weight,
|
||
effective_weight, weighted_score}. Exactly what result.indicators
|
||
carries minus the weight-0 ATR entry."""
|
||
i, f = row.i, row.f
|
||
out: Dict[str, Dict] = {}
|
||
if not i["valid"]:
|
||
return out
|
||
if i["ema_present"]:
|
||
if ema_mode_2ema:
|
||
desc = _EMA2_DESC[i["ema2_desc"]]
|
||
desc += {1: " | Golden cross (50/200)", 2: " | Death cross (50/200)"}.get(i["ema_cross"], "")
|
||
value = 0 if i["ema2_desc"] == 0 else f["ema_value"]
|
||
else:
|
||
cross = {1: " | Golden cross (fast/medium)", 2: " | Death cross (fast/medium)"}.get(i["ema_cross"], "")
|
||
ribbon = {1: " | Ribbon COMPRESSED (trend change imminent)",
|
||
2: " | Ribbon EXPANDING (new trend starting)"}.get(i["ema_ribbon"], "")
|
||
desc = f"EMA alignment: {i['ema_align']}/{i['ema_max_checks']}{cross}{ribbon}"
|
||
value = f["ema_value"]
|
||
out["ema_trend"] = dict(value=value, signal=i["ema_signal"], confidence=f["ema_conf"], description=desc,
|
||
weight=1.0, effective_weight=f["ew_ema"], weighted_score=f["ws_ema"])
|
||
if i["rsi_present"]:
|
||
v = f["rsi_value"]
|
||
out["rsi"] = dict(value=v, signal=i["rsi_signal"], confidence=f["rsi_conf"],
|
||
description=f"RSI {v:.1f} — {_RSI_SUFFIX[i['rsi_desc']]}",
|
||
weight=1.5, effective_weight=f["ew_rsi"], weighted_score=f["ws_rsi"])
|
||
if i["bb_present"]:
|
||
v = f["bb_value"]
|
||
note = " | SQUEEZE detected (expect breakout)" if i["bb_squeeze"] else ""
|
||
out["bollinger"] = dict(value=v, signal=i["bb_signal"], confidence=f["bb_conf"],
|
||
description=f"{_BB_DESC[i['bb_desc']]} ({v:.2f}){note}",
|
||
weight=1.5, effective_weight=f["ew_bb"], weighted_score=f["ws_bb"])
|
||
if i["adx_present"]:
|
||
slope = f["adx_slope"]
|
||
sd = {1: " | TREND EMERGING (ADX rising fast)",
|
||
2: f" | Trend strengthening (slope +{slope:.1f})",
|
||
3: f" | Trend weakening (slope {slope:.1f})"}.get(i["adx_slope_desc"], "")
|
||
desc = (f"ADX {f['adx_value']:.1f} ({_ADX_STRENGTH[i['adx_strength']]} {_ADX_DIR[i['adx_dir']]}) | "
|
||
f"+DI={f['plus_di']:.1f} -DI={f['minus_di']:.1f}{sd}")
|
||
out["adx"] = dict(value=f["adx_value"], signal=i["adx_signal"], confidence=f["adx_conf"], description=desc,
|
||
weight=0.5, effective_weight=f["ew_adx"], weighted_score=f["ws_adx"])
|
||
if i["macd_present"]:
|
||
out["macd"] = dict(value=f["macd_value"], signal=i["macd_signal"], confidence=f["macd_conf"],
|
||
description=_MACD_DESC[i["macd_desc"]],
|
||
weight=1.5, effective_weight=f["ew_macd"], weighted_score=f["ws_macd"])
|
||
if i["roc_present"]:
|
||
v = f["roc_value"]
|
||
d = i["roc_desc"]
|
||
desc = {0: f"ROC +{v:.2f}% — Strong momentum acceleration", 1: f"ROC +{v:.2f}% — Rising momentum",
|
||
2: f"ROC {v:.2f}% — Strong bearish momentum", 3: f"ROC {v:.2f}% — Falling momentum",
|
||
4: f"ROC {v:+.2f}% — No clear momentum"}[d]
|
||
out["roc"] = dict(value=v, signal=i["roc_signal"], confidence=f["roc_conf"], description=desc,
|
||
weight=1.5, effective_weight=f["ew_roc"], weighted_score=f["ws_roc"])
|
||
if i["kc_present"]:
|
||
out["keltner"] = dict(value=f["kc_value"], signal=i["kc_signal"], confidence=f["kc_conf"],
|
||
description=_KC_DESC[i["kc_desc"]],
|
||
weight=1.5, effective_weight=f["ew_kc"], weighted_score=f["ws_kc"])
|
||
if i["div_present"]:
|
||
items = [s for b, s in enumerate(_DIV_ITEMS) if i["div_mask"] & (1 << b)]
|
||
out["divergence"] = dict(value=len(items), signal=i["div_signal"], confidence=70,
|
||
description=" | ".join(items),
|
||
weight=1.5, effective_weight=f["ew_div"], weighted_score=f["ws_div"])
|
||
if i["cs_present"]:
|
||
items = [s for b, s in enumerate(_CS_ITEMS) if i["cs_mask"] & (1 << b)]
|
||
out["candlestick"] = dict(value=f["cs_value"], signal=i["cs_signal"], confidence=f["cs_conf"],
|
||
description=" | ".join(items),
|
||
weight=1.0, effective_weight=f["ew_cs"], weighted_score=f["ws_cs"])
|
||
return out
|
||
|
||
|
||
def pattern_signals(row: K1Row) -> List[str]:
|
||
if not row.i["valid"] or not row.i["cs_present"]:
|
||
return []
|
||
return [s for b, s in enumerate(_CS_ITEMS) if row.i["cs_mask"] & (1 << b)]
|
||
|
||
|
||
def key_levels(row: K1Row, ema_mode_2ema: bool) -> Dict[str, float]:
|
||
"""result.key_levels in insertion order."""
|
||
i, f = row.i, row.f
|
||
kl: Dict[str, float] = {}
|
||
if not i["valid"]:
|
||
return kl
|
||
if i["atr_present"]:
|
||
kl["atr"] = f["atr"]
|
||
kl["atr_pct"] = f["atr_pct"]
|
||
if i["ema_present"]:
|
||
if i["ema_trend_present"]:
|
||
kl["ema_trend"] = f["ema_trend"]
|
||
if not ema_mode_2ema:
|
||
kl["ema_fast"] = f["ema_fast"]
|
||
kl["ema_medium"] = f["ema_medium"]
|
||
kl["ema_slow"] = f["ema_slow"]
|
||
if i["bb_present"]:
|
||
kl["bb_upper"] = f["bb_upper"]; kl["bb_lower"] = f["bb_lower"]; kl["bb_mid"] = f["bb_mid"]
|
||
if i["kc_present"]:
|
||
kl["kc_upper"] = f["kc_upper"]; kl["kc_lower"] = f["kc_lower"]; kl["kc_mid"] = f["kc_mid"]
|
||
if not i["sh_fallback"]:
|
||
for a in range(3):
|
||
if i["sh_mask"] & (1 << a):
|
||
kl[f"swing_high_{a + 1}"] = f[f"sh{a + 1}"]
|
||
if not i["sl_fallback"]:
|
||
for a in range(3):
|
||
if i["sl_mask"] & (1 << a):
|
||
kl[f"swing_low_{a + 1}"] = f[f"sl{a + 1}"]
|
||
if i["sh_fallback"]:
|
||
kl["swing_high_1"] = f["sh1"]
|
||
if i["sl_fallback"]:
|
||
kl["swing_low_1"] = f["sl1"]
|
||
return kl
|
||
|
||
|
||
def format_warning(code: int, arg: float, ctx: Dict) -> str:
|
||
"""signal_aggregator's warning text for one K2 warning code."""
|
||
n = ck.WARN_NAMES[code]
|
||
if n == "conflict":
|
||
return "Timeframes CONFLICT — mixed signals across timeframes"
|
||
if n == "counter_trend":
|
||
return "Counter-trend signal — reduced confidence"
|
||
if n == "extreme_vol":
|
||
return "EXTREME volatility — wider stops needed"
|
||
if n == "low_vol":
|
||
return "Low volatility — potential for breakout"
|
||
if n == "ema200_near":
|
||
return f"NEAR EMA200 ({arg:.2f}%) — chop risk"
|
||
if n == "ema200_far":
|
||
return f"FAR FROM EMA200 ({arg:.2f}%) — over-extended"
|
||
if n == "exhaustion_low":
|
||
return "TREND EXHAUSTION — price extended >3.5 ATR past swing low"
|
||
if n == "exhaustion_high":
|
||
return "TREND EXHAUSTION — price extended >3.5 ATR past swing high"
|
||
if n == "no_pullback":
|
||
return f"NO PULLBACK — price {arg:.1f} ATR from EMA21"
|
||
if n == "weak_pullback":
|
||
return f"Weak pullback — price {arg:.1f} ATR from EMA21"
|
||
if n == "rsi_ext_up":
|
||
return f"Extended — RSI {arg:.0f} in uptrend"
|
||
if n == "rsi_ext_down":
|
||
return f"Extended — RSI {arg:.0f} in downtrend"
|
||
if n == "exhaustion_block":
|
||
return f"EXHAUSTION BLOCK — {int(arg)} exhaustion signals fired, trade rejected"
|
||
if n == "session_filter":
|
||
return f"SESSION FILTER: {int(arg):02}:00 UTC blocked"
|
||
if n == "day_filter":
|
||
day_names = {0: "Mon", 1: "Tue", 2: "Wed", 3: "Thu", 4: "Fri", 5: "Sat", 6: "Sun"}
|
||
return f"DAY FILTER: {day_names.get(int(arg), '?')} not in allowed days"
|
||
if n == "exhausted_trend":
|
||
return "EXHAUSTED TREND FILTER: C-grade blocked in exhausted trend"
|
||
if n == "sell_quality":
|
||
return f"SELL FILTER: quality {ck.QUALITY_NAMES[int(arg)]} below minimum for SELL"
|
||
if n == "mtf_opposition":
|
||
return "MTF OPPOSITION: majority of timeframes point against signal — no trade"
|
||
if n == "sell_vol":
|
||
return f"SELL FILTER: volatility '{ck.VOL_NAMES[int(arg)]}' too quiet for shorts"
|
||
if n == "regime_blocked":
|
||
return f"REGIME FILTER: '{ck.REGIME_NAMES[int(arg)]}' blocked for XAU"
|
||
if n == "sell_regime":
|
||
return f"SELL FILTER: regime '{ck.REGIME_NAMES[int(arg)]}' blocked for shorts"
|
||
if n == "sell_hour":
|
||
return f"SELL FILTER: {int(arg):02}:00 UTC entry blocked for shorts"
|
||
if n == "big_candle":
|
||
return (f"BIG CANDLE FILTER: {arg:.1f}x ATR bar at {ctx['bc_ts']} within "
|
||
f"{ctx['big_candle_lookback_hours']:g}h — no entries")
|
||
if n == "m5_score":
|
||
return f"M5 FILTER: |score| {arg:.1f} < {ctx['m5_min_abs_score']:.0f} minimum"
|
||
if n == "m5_mtf":
|
||
return "M5 FILTER: no MTF agreement — M5 signals require confirmation"
|
||
if n == "m5_conviction":
|
||
return "M5 FILTER: no confirmation TF shows directional conviction"
|
||
if n == "m5_sell_quality":
|
||
return (f"M5 FILTER: SELL quality {ck.QUALITY_NAMES[int(arg)]} capped — "
|
||
f"high-conviction M5 shorts enter at extremes")
|
||
if n == "m5_vol":
|
||
return (f"M5 FILTER: '{ck.VOL_NAMES[int(arg)]}' volatility — "
|
||
f"chop-transition regime, stops get swept")
|
||
if n == "m5_sell_rsi":
|
||
return (f"M5 FILTER: SELL blocked — RSI {arg:.0f} < "
|
||
f"{ctx['m5_sell_min_rsi']:.0f}, dip already extended")
|
||
if n == "spread_gate_atr" or n == "spread_gate_generic":
|
||
return (f"SPREAD GATE: target ${arg:.2f} < ${ctx['min_tp']:.2f} minimum — "
|
||
f"costs would eat the edge")
|
||
if n == "low_atr":
|
||
return f"LOW ATR FILTER: ATR ${arg:.2f} < ${ctx['min_atr_threshold']:.2f} minimum"
|
||
if n == "rf_unavailable":
|
||
return f"Regime filter requested but {ctx['rf_tf']} data unavailable"
|
||
if n == "rf_no_adx":
|
||
return f"Regime filter: ADX could not be computed on {ctx['rf_tf']}"
|
||
if n == "rf_blocked":
|
||
return (f"REGIME FILTER: {ctx['rf_tf']} ADX={arg:.1f} < "
|
||
f"{ctx['regime_filter_min_adx']:.0f} — market choppy, skipping signal")
|
||
raise ValueError(f"unknown warning code {code}")
|
||
|
||
|
||
def collect_reasons(inds: Dict[str, Dict]) -> Tuple[List[str], List[str]]:
|
||
"""signal_aggregator._collect_reasons (news never contributes here)."""
|
||
bullish: List[str] = []
|
||
bearish: List[str] = []
|
||
for name, ind in inds.items():
|
||
sig = ind["signal"]
|
||
label = ck.INDICATOR_LABELS[name]
|
||
if sig > 0:
|
||
prefix = "STRONG: " if sig == 2 else ""
|
||
bullish.append(f"{prefix}{label}: {ind['description']}")
|
||
elif sig < 0:
|
||
prefix = "STRONG: " if sig == -2 else ""
|
||
bearish.append(f"{prefix}{label}: {ind['description']}")
|
||
return bullish, bearish
|
||
|
||
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
# Host flags: the calendar / session rules of _replay_span, vectorised
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
|
||
def host_flags(primary_index: pd.DatetimeIndex, platform_cfg: PlatformConfig,
|
||
bt_cfg: BacktestConfig) -> Dict[str, np.ndarray]:
|
||
"""Per primary bar: the replay skip reason (1 WEEKEND_FILTER, 2 DAY_FILTER,
|
||
3 SESSION_WINDOW, 0 analysed), the bar-label hour/weekday (gate 6b) and
|
||
the bar-close hour/weekday/minute/day (gate 6g3, XAU_CP, features)."""
|
||
n = len(primary_index)
|
||
primary_dur = pd.Timedelta(seconds=BAR_SECONDS[bt_cfg.primary_tf])
|
||
decision = primary_index + primary_dur
|
||
skip = np.zeros(n, dtype=np.int8)
|
||
is_xau = PAIR_INFO[bt_cfg.pair]["display"] == "XAU/USD"
|
||
xau = platform_cfg.xau_strategy
|
||
if is_xau:
|
||
if xau.weekend_filter:
|
||
ny = decision.tz_convert(_NY_TZ)
|
||
wd = np.asarray(ny.weekday)
|
||
hr = np.asarray(ny.hour)
|
||
weekend = ((wd == 4) & (hr >= 17)) | (wd == 5) | ((wd == 6) & (hr < 18))
|
||
skip[weekend] = 1
|
||
if xau.allowed_weekdays:
|
||
day = ~np.isin(np.asarray(decision.weekday), list(xau.allowed_weekdays))
|
||
skip[day & (skip == 0)] = 2
|
||
flt = platform_cfg.backtest
|
||
if flt.session_filter:
|
||
h = np.asarray(decision.hour)
|
||
blocked = xau.blocked_hours_utc if is_xau else flt.blocked_hours_utc
|
||
sess = ~((h >= flt.session_start_hour) & (h < flt.session_end_hour)) | np.isin(h, list(blocked))
|
||
skip[sess & (skip == 0)] = 3
|
||
return {
|
||
"skip_reason": skip,
|
||
"label_hour": np.asarray(primary_index.hour, dtype=np.int8),
|
||
"label_wd": np.asarray(primary_index.weekday, dtype=np.int8),
|
||
"close_hour": np.asarray(decision.hour, dtype=np.int8),
|
||
"close_wd": np.asarray(decision.weekday, dtype=np.int8),
|
||
"close_minute": np.asarray(decision.minute, dtype=np.int8),
|
||
"close_day": np.asarray(decision.day, dtype=np.int8),
|
||
}
|
||
|
||
|
||
def replay_bounds(primary_df: pd.DataFrame, bt_cfg: BacktestConfig) -> Tuple[int, int]:
|
||
bt_start_ts = pd.Timestamp(bt_cfg.start)
|
||
bt_end_ts = pd.Timestamp(bt_cfg.end)
|
||
first_eligible_pos = max(bt_cfg.warmup_bars, int(primary_df.index.searchsorted(bt_start_ts)))
|
||
last_pos = int(primary_df.index.searchsorted(bt_end_ts, side="right"))
|
||
last_pos = min(last_pos, len(primary_df))
|
||
return first_eligible_pos, last_pos
|
||
|
||
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
# The pipeline
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
|
||
class _Run:
|
||
"""Everything one replay produces on the device plus the host handles
|
||
the builders need."""
|
||
|
||
|
||
def refuse_unsupported(platform_cfg: PlatformConfig, bt_cfg: BacktestConfig) -> None:
|
||
t = platform_cfg.technical
|
||
x = platform_cfg.xau_strategy
|
||
problems = []
|
||
if len(platform_cfg.confirmation_timeframes) > ck.MAX_CONFIRM:
|
||
problems.append(f"{len(platform_cfg.confirmation_timeframes)} confirmation timeframes "
|
||
f"(the CUDA aggregate gathers at most {ck.MAX_CONFIRM})")
|
||
if t.divergence_lookback >= ck.MAX_DIV_LOOKBACK:
|
||
problems.append(f"technical.divergence_lookback={t.divergence_lookback} "
|
||
f"(CUDA limit {ck.MAX_DIV_LOOKBACK - 1})")
|
||
if max(t.swing_sr_lookback, 120) >= ck.MAX_SR_LOOKBACK:
|
||
problems.append(f"technical.swing_sr_lookback={t.swing_sr_lookback} "
|
||
f"(CUDA limit {ck.MAX_SR_LOOKBACK - 1})")
|
||
if x.cp_lookback + 1 > ck.MAX_CP_LOOKBACK:
|
||
problems.append(f"xau_strategy.cp_lookback={x.cp_lookback} (CUDA limit {ck.MAX_CP_LOOKBACK - 1})")
|
||
if x.big_candle_atr_multiple > 0:
|
||
bar_secs = BaseDataProvider.CANDLE_SECONDS[platform_cfg.primary_timeframe]
|
||
n_bc = max(1, int(x.big_candle_lookback_hours * 3600 // bar_secs))
|
||
if n_bc + 3 * t.atr_period > ck.FEAT_MAX_P:
|
||
problems.append(f"big-candle window {n_bc + 3 * t.atr_period} bars (CUDA limit {ck.FEAT_MAX_P})")
|
||
if t.ema_mode not in ("4ema", "2ema"):
|
||
problems.append(f"technical.ema_mode={t.ema_mode!r}")
|
||
if problems:
|
||
raise SystemExit("cudaBasedBacktesting.py cannot run this configuration: "
|
||
+ "; ".join(problems) + ". Use backtest/gpu_backtest.py.")
|
||
|
||
|
||
def analyze_frames(bt_cfg: BacktestConfig, platform_cfg: PlatformConfig,
|
||
dfs_full: Dict[Timeframe, pd.DataFrame], timeframes: Sequence[Timeframe]) -> _Run:
|
||
"""Precompute -> K1 per timeframe -> alignment -> host flags -> K2 ->
|
||
cooldown, on already loaded (cleaned) frames. Returns the run context
|
||
the builders read. Shared by run_backtest_cuda and the parity tests."""
|
||
cp = _gk._require_cp()
|
||
flt = platform_cfg.backtest
|
||
t0 = time.perf_counter()
|
||
caches = {tf: precompute_indicators(dfs_full[tf], platform_cfg.technical, gpu_log=True) for tf in timeframes}
|
||
logger.info(f"Indicator pre-compute took {time.perf_counter() - t0:.2f}s (CuPy GPU)")
|
||
|
||
pf, pi, rm = ck.pack_params(platform_cfg, bt_cfg)
|
||
t1 = time.perf_counter()
|
||
dev = {tf: ck.upload_cache(caches[tf], dfs_full[tf]) for tf in timeframes}
|
||
an = {tf: ck.analyze_tf(dev[tf], pf, pi, rm, tf.value == "5min") for tf in timeframes}
|
||
logger.info(f"K1: {sum(dev[tf]['n'] for tf in timeframes)} bars of "
|
||
f"{len(timeframes)} timeframes analysed on the device in {time.perf_counter() - t1:.3f}s")
|
||
|
||
primary_df = dfs_full[bt_cfg.primary_tf]
|
||
primary_dev = dev[bt_cfg.primary_tf]
|
||
primary_secs = BAR_SECONDS[bt_cfg.primary_tf]
|
||
cuts = {tf: ck.align_cut(dev[tf]["index_ns"], primary_dev["index_ns"], primary_secs, BAR_SECONDS[tf])
|
||
for tf in timeframes if tf != bt_cfg.primary_tf}
|
||
confirms = [(an[tf], cuts[tf]) for tf in bt_cfg.confirmation_tfs]
|
||
rf = None
|
||
if flt.regime_filter:
|
||
rtf = flt.regime_filter_timeframe
|
||
rf_cut = cuts[rtf] if rtf != bt_cfg.primary_tf else cp.arange(1, primary_dev["n"] + 1, dtype=cp.int64)
|
||
rf = (dev[rtf]["adx"], rf_cut)
|
||
flags = host_flags(primary_df.index, platform_cfg, bt_cfg)
|
||
first_eligible_pos, last_pos = replay_bounds(primary_df, bt_cfg)
|
||
logger.info(f"Replaying {bt_cfg.pair} {bt_cfg.primary_tf.value}: "
|
||
f"bars {first_eligible_pos}–{last_pos} of {len(primary_df)}")
|
||
n_bars = last_pos - first_eligible_pos
|
||
|
||
t_loop = time.perf_counter()
|
||
k2 = ck.aggregate(primary_dev, an[bt_cfg.primary_tf], confirms, rf, flags, pf, pi,
|
||
first_eligible_pos, last_pos)
|
||
logger.info(f"K2: {n_bars} primary bars aggregated on the device in {time.perf_counter() - t_loop:.3f}s")
|
||
logger.info("GPU cache fallbacks: 0 (no per-bar recompute path in cudaBasedBacktesting)")
|
||
|
||
status = k2.i("status").get()[:n_bars] if n_bars > 0 else np.zeros(0, dtype=np.int32)
|
||
cand_pos = (np.flatnonzero(status == 2) + first_eligible_pos).astype(np.int64)
|
||
accepted_mask = ck.cooldown(cand_pos, bt_cfg.cooldown_bars)
|
||
acc_pos = cand_pos[accepted_mask.astype(bool)]
|
||
|
||
skipped_reasons: Dict[str, int] = {}
|
||
skip = flags["skip_reason"][first_eligible_pos:last_pos]
|
||
order: List[str] = []
|
||
for code in skip:
|
||
if code:
|
||
name = ck.SKIP_NAMES[int(code)]
|
||
if name not in skipped_reasons:
|
||
order.append(name)
|
||
skipped_reasons[name] = skipped_reasons.get(name, 0) + 1
|
||
skipped_reasons = {name: skipped_reasons[name] for name in order}
|
||
|
||
ctx = _Run()
|
||
ctx.bt_cfg, ctx.platform_cfg, ctx.timeframes = bt_cfg, platform_cfg, list(timeframes)
|
||
ctx.dfs_full, ctx.caches, ctx.dev, ctx.an, ctx.cuts, ctx.k2, ctx.flags = dfs_full, caches, dev, an, cuts, k2, flags
|
||
ctx.pf, ctx.pi, ctx.rm = pf, pi, rm
|
||
ctx.first_pos, ctx.last_pos, ctx.n_bars = first_eligible_pos, last_pos, n_bars
|
||
ctx.status, ctx.cand_pos, ctx.acc_pos, ctx.skipped_reasons = status, cand_pos, acc_pos, skipped_reasons
|
||
ctx.t_loop = t_loop
|
||
return ctx
|
||
|
||
|
||
def run_backtest_cuda(bt_cfg: BacktestConfig, platform_cfg: PlatformConfig) -> Tuple[List[Trade], Dict[Timeframe, pd.DataFrame]]:
|
||
"""gpu_backtest.run_backtest, with the replay on the device."""
|
||
cp = _gk._require_cp()
|
||
if platform_cfg.primary_timeframe != bt_cfg.primary_tf:
|
||
raise ValueError(
|
||
f"primary_tf mismatch: BacktestConfig={bt_cfg.primary_tf.value} "
|
||
f"vs PlatformConfig={platform_cfg.primary_timeframe.value}")
|
||
refuse_unsupported(platform_cfg, bt_cfg)
|
||
provider = make_data_provider(platform_cfg)
|
||
|
||
# ── Load all required timeframes (same files as the reference script) ──
|
||
timeframes = [bt_cfg.primary_tf] + list(bt_cfg.confirmation_tfs)
|
||
flt = platform_cfg.backtest
|
||
if flt.regime_filter and flt.regime_filter_timeframe not in timeframes:
|
||
timeframes.append(flt.regime_filter_timeframe)
|
||
dfs_full: Dict[Timeframe, pd.DataFrame] = {}
|
||
cache_names: Dict[str, str] = {}
|
||
tf_starts = {}
|
||
from datetime import timedelta
|
||
for tf in timeframes:
|
||
tf_start = bt_cfg.start - timedelta(days=_history_pad_days(tf, bt_cfg.warmup_bars))
|
||
tf_starts[tf] = tf_start
|
||
cache_names[tf.value] = _cache_path(bt_cfg.pair, tf, tf_start, bt_cfg.end, provider.name).name
|
||
df = fetch_history(provider, bt_cfg.pair, tf, start=tf_start, end=bt_cfg.end,
|
||
use_cache=bt_cfg.use_cache)
|
||
if df.empty:
|
||
raise RuntimeError(f"No data for {bt_cfg.pair} {tf.value}")
|
||
ohlc = df[["Open", "High", "Low", "Close"]].to_numpy(dtype=np.float64)
|
||
if not np.isfinite(ohlc).all():
|
||
raise RuntimeError(f"{tf.value}: the cleaned frame carries non-finite OHLC — "
|
||
f"the device replay assumes clean bars")
|
||
dfs_full[tf] = df
|
||
|
||
ctx = analyze_frames(bt_cfg, platform_cfg, dfs_full, timeframes)
|
||
primary_df = dfs_full[bt_cfg.primary_tf]
|
||
first_eligible_pos, last_pos, n_bars = ctx.first_pos, ctx.last_pos, ctx.n_bars
|
||
acc_pos, cand_pos, skipped_reasons, t_loop = ctx.acc_pos, ctx.cand_pos, ctx.skipped_reasons, ctx.t_loop
|
||
trades = build_trades(ctx, acc_pos)
|
||
signal_ts_list = [primary_df.index[int(p)] for p in acc_pos]
|
||
|
||
replay_seconds = time.perf_counter() - t_loop
|
||
logger.info(
|
||
f"Replay complete: {len(trades)} entries in {replay_seconds:.1f}s "
|
||
f"({n_bars / max(replay_seconds, 1e-6):.0f} bars/s, device)")
|
||
if skipped_reasons:
|
||
logger.info(f"Skipped bars by reason: {skipped_reasons}")
|
||
|
||
# ── Resolve trade outcomes on the RAW price path (shared GPU resolver) ──
|
||
sim_df = fetch_history(provider, bt_cfg.pair, bt_cfg.primary_tf, start=tf_starts[bt_cfg.primary_tf],
|
||
end=bt_cfg.end, use_cache=bt_cfg.use_cache, clean=False)
|
||
_reanchor_entries_to_raw(trades, signal_ts_list, sim_df)
|
||
if bt_cfg.fixed_rr_tp > 0:
|
||
_apply_fixed_rr_tp(trades, bt_cfg.fixed_rr_tp)
|
||
logger.info(
|
||
f"RESEARCH override: fixed-RR TP = {bt_cfg.fixed_rr_tp} x risk; "
|
||
f"trailing exit DISABLED for all {len(trades)} trades (incl. XAU_PB)")
|
||
run_arm = 0.0 if bt_cfg.fixed_rr_tp > 0 else platform_cfg.signals.trail_arm_r
|
||
t_sim = time.perf_counter()
|
||
simulate_trades(trades, sim_df, max_bars=bt_cfg.max_hold_bars, spread=bt_cfg.spread,
|
||
trail_arm_r=run_arm, trail_dist_r=platform_cfg.signals.trail_dist_r)
|
||
logger.info(f"Trade simulation took {time.perf_counter() - t_sim:.2f}s (CuPy GPU)")
|
||
|
||
# ── research flags ──
|
||
if bt_cfg.export_features and trades:
|
||
t_f = time.perf_counter()
|
||
attach_features(ctx, acc_pos, trades)
|
||
logger.info(f"K3: {len(trades)} feature vectors on the device in {time.perf_counter() - t_f:.3f}s")
|
||
if bt_cfg.record_dir:
|
||
t_r = time.perf_counter()
|
||
attach_snapshots(ctx, acc_pos, trades)
|
||
bar_chunks = build_bar_chunks(ctx) if bt_cfg.record_bars else None
|
||
logger.info(f"--record: snapshots{' + bar rows' if bt_cfg.record_bars else ''} built in "
|
||
f"{time.perf_counter() - t_r:.2f}s")
|
||
replay_record.write_run(
|
||
bt_cfg.record_dir, trades=trades, book_df=trades_to_dataframe(trades),
|
||
bt_cfg=bt_cfg, platform_cfg=platform_cfg, aggregator=SignalAggregator(platform_cfg),
|
||
run_info=dict(first_pos=first_eligible_pos, last_pos=last_pos,
|
||
n_primary_bars=len(primary_df), n_bars_in_window=n_bars,
|
||
skipped=skipped_reasons, n_candidates=int(len(cand_pos)),
|
||
workers=0, backend=BACKEND.describe(),
|
||
cache_files=cache_names, replay_seconds=replay_seconds,
|
||
cache_fallbacks=0),
|
||
sim_df=sim_df, bar_chunks=bar_chunks)
|
||
logger.info(f"--record: wrote run to {bt_cfg.record_dir}")
|
||
return trades, dfs_full
|
||
|
||
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
# Host builders: Trade objects, warnings, snapshots, features, bar rows
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
|
||
def _k2_rows(k2: "ck.K2Out", positions: np.ndarray):
|
||
"""K2 fields + warning codes/args for the given absolute bar positions."""
|
||
idx = np.asarray(positions, dtype=np.int64) - k2.first_pos
|
||
if len(idx) == 0:
|
||
return [], [], np.zeros((0, ck.MAX_WARN), np.int16), np.zeros((0, ck.MAX_WARN), np.float64)
|
||
fi = k2.OF2[:, idx].get()
|
||
ii = k2.OI2[:, idx].get()
|
||
wc = k2.warn_code[idx].get()
|
||
wa = k2.warn_arg[idx].get()
|
||
ints = [{n: int(ii[k, m]) for k, n in enumerate(ck.K2_INT_FIELDS)} for m in range(len(idx))]
|
||
floats = [{n: float(fi[k, m]) for k, n in enumerate(ck.K2_FLOAT_FIELDS)} for m in range(len(idx))]
|
||
return ints, floats, wc, wa
|
||
|
||
|
||
def _warning_ctx(ctx, pos: int, r2i: Dict, r2f: Dict) -> Dict:
|
||
s = ctx.platform_cfg.signals
|
||
x = ctx.platform_cfg.xau_strategy
|
||
primary_df = ctx.dfs_full[ctx.bt_cfg.primary_tf]
|
||
return {
|
||
"min_tp": s.round_trip_cost * s.min_tp_cost_multiple,
|
||
"min_atr_threshold": x.min_atr_threshold,
|
||
"m5_min_abs_score": x.m5_min_abs_score,
|
||
"m5_sell_min_rsi": x.m5_sell_min_rsi,
|
||
"big_candle_lookback_hours": x.big_candle_lookback_hours,
|
||
"bc_ts": primary_df.index[r2i["bc_idx"]] if r2i["bc_idx"] >= 0 else None,
|
||
"rf_tf": ctx.platform_cfg.backtest.regime_filter_timeframe.value,
|
||
"regime_filter_min_adx": ctx.platform_cfg.backtest.regime_filter_min_adx,
|
||
}
|
||
|
||
|
||
_RF_CODES = {ck.WARN_CODES["rf_unavailable"], ck.WARN_CODES["rf_no_adx"], ck.WARN_CODES["rf_blocked"]}
|
||
|
||
|
||
def build_warnings(ctx, pos: int, r2i: Dict, r2f: Dict, codes, args) -> Tuple[List[str], List[str]]:
|
||
"""(warnings, cleaned_warnings) of one analysed bar, after the XAU_CP
|
||
cleanup (when a pattern setup was accepted) and the stale spread-gate
|
||
step (when a setup existed at step 11), with the regime-filter lines
|
||
appended last like the replay does."""
|
||
wctx = _warning_ctx(ctx, pos, r2i, r2f)
|
||
n = r2i["warn_n"]
|
||
main: List[str] = []
|
||
rf: List[str] = []
|
||
for q in range(n):
|
||
code = int(codes[q])
|
||
text = format_warning(code, float(args[q]), wctx)
|
||
(rf if code in _RF_CODES else main).append(text)
|
||
ns = types.SimpleNamespace(warnings=main, cleaned_warnings=[],
|
||
trade_setup=object() if r2i["strategy_pre_rf"] != 0 else None)
|
||
if r2i["cp_accepted"]:
|
||
SignalAggregator._clean_accepted_cp_warnings(ns)
|
||
SignalAggregator._drop_stale_spread_gate_vetoes(ns)
|
||
return list(ns.warnings) + rf, list(ns.cleaned_warnings)
|
||
|
||
|
||
def build_reasons(ctx, r2i: Dict, r2f: Dict, inds: Dict[str, Dict]) -> Tuple[List[str], List[str]]:
|
||
bullish, bearish = collect_reasons(inds)
|
||
strat = r2i["strategy"] if r2i["strategy"] else r2i["strategy_pre_rf"]
|
||
if r2i["pb_accepted"]:
|
||
marker = (f"XAU PULLBACK STRATEGY: retrace {r2f['pb_retrace_atr']:.1f} "
|
||
f"ATR to EMA21/KC-mid, trend resumed")
|
||
(bullish if r2i["direction"] > 0 else bearish).insert(0, marker)
|
||
if r2i["cp_accepted"]:
|
||
atr = ctx_atr = None
|
||
pattern = ck.PATTERN_ORDER[r2i["cp_pattern"]]
|
||
direction = "BUY" if r2i["direction"] > 0 else "SELL"
|
||
marker = (f"XAU CHART PATTERN: {pattern} ({direction}) break of {r2f['cp_trigger']:.2f}, "
|
||
f"height {r2f['cp_height'] / r2f['_atr']:.1f} ATR")
|
||
(bullish if r2i["direction"] > 0 else bearish).insert(0, marker)
|
||
if r2i["rf_state"] == 4:
|
||
rtf = ctx.platform_cfg.backtest.regime_filter_timeframe.value
|
||
line = f"Regime: {rtf} ADX={r2f['rf_adx']:.1f} confirms trending market"
|
||
(bullish if r2f["final_score"] > 0 else bearish).append(line)
|
||
return bullish, bearish
|
||
|
||
|
||
def build_trades(ctx, acc_pos: np.ndarray) -> List[Trade]:
|
||
bt_cfg, platform_cfg = ctx.bt_cfg, ctx.platform_cfg
|
||
primary_df = ctx.dfs_full[bt_cfg.primary_tf]
|
||
pan = ctx.an[bt_cfg.primary_tf]
|
||
two = platform_cfg.technical.ema_mode == "2ema"
|
||
ints, floats, wc, wa = _k2_rows(ctx.k2, acc_pos)
|
||
prows = k1_rows(pan, acc_pos)
|
||
pip_value = PAIR_INFO[bt_cfg.pair]["pip_value"]
|
||
trades: List[Trade] = []
|
||
for m, pos in enumerate(acc_pos):
|
||
pos = int(pos)
|
||
r2i, r2f, prow = ints[m], floats[m], prows[m]
|
||
r2f["_atr"] = prow.f["atr"] if prow.i["atr_present"] else 0.0
|
||
direction = "BUY" if r2i["direction"] > 0 else "SELL"
|
||
tag = ck.STRATEGY_NAMES[r2i["strategy"]] or "GENERIC"
|
||
inds = describe_indicators(prow, two)
|
||
# gpu_backtest._build_indicator_breakdown
|
||
breakdown: Dict[str, Dict] = {}
|
||
agree = oppose = neutral = 0
|
||
is_buy = direction == "BUY"
|
||
for name, ind in inds.items():
|
||
sig = ind["signal"]
|
||
agrees = "neutral"
|
||
if sig > 0:
|
||
agrees = "agree" if is_buy else "oppose"
|
||
elif sig < 0:
|
||
agrees = "agree" if not is_buy else "oppose"
|
||
breakdown[name] = {
|
||
"value": float(ind["value"]) if ind["value"] is not None else None,
|
||
"signal": ck.SIGNAL_NAMES[sig],
|
||
"confidence": float(ind["confidence"]),
|
||
"weight": float(ind["weight"]),
|
||
"agrees": agrees,
|
||
"description": ind["description"],
|
||
}
|
||
if agrees == "agree":
|
||
agree += 1
|
||
elif agrees == "oppose":
|
||
oppose += 1
|
||
else:
|
||
neutral += 1
|
||
total = len(breakdown)
|
||
bullish, bearish = build_reasons(ctx, r2i, r2f, inds)
|
||
reasons = bullish if direction == "BUY" else bearish
|
||
warnings, _cleaned = build_warnings(ctx, pos, r2i, r2f, wc[m], wa[m])
|
||
# the setup's own pip value: 0.10 hard-coded by the XAU ATR generator, PAIR_INFO otherwise
|
||
setup_pip = 0.10 if tag == "XAU_ATR" else pip_value
|
||
# the setup's own trailing parameters (XAU_PB carries the pullback pair)
|
||
trail_arm = platform_cfg.signals.pb_trail_arm_r if tag == "XAU_PB" else 0.0
|
||
trail_dist = platform_cfg.signals.pb_trail_dist_r if tag == "XAU_PB" else 0.0
|
||
trades.append(Trade(
|
||
pair=bt_cfg.pair,
|
||
entry_time=primary_df.index[pos + 1],
|
||
entry_price=r2f["entry_price"],
|
||
sl=r2f["adj_sl"], tp1=r2f["adj_tp1"], tp2=r2f["adj_tp2"],
|
||
direction=direction,
|
||
pip_value=setup_pip,
|
||
risk_pips=r2f["sl_dist"] / setup_pip,
|
||
reward_pips=r2f["tp1_dist"] / setup_pip,
|
||
rr_planned=r2f["rr_rounded"],
|
||
quality=ck.QUALITY_NAMES[r2i["quality"]],
|
||
final_score=r2f["final_score"],
|
||
final_confidence=r2f["final_conf"],
|
||
market_regime=ck.REGIME_NAMES[prow.i["regime"]],
|
||
volatility_regime=ck.VOL_NAMES[prow.i["vol"]],
|
||
strategy_tag=tag,
|
||
entry_reasons=list(reasons),
|
||
entry_warnings=list(warnings),
|
||
indicator_breakdown=breakdown,
|
||
indicators_agreeing=agree, indicators_opposing=oppose,
|
||
indicators_neutral=neutral, indicators_total=total,
|
||
stake_multiplier=1.0,
|
||
trail_arm_r=trail_arm, trail_dist_r=trail_dist,
|
||
))
|
||
return trades
|
||
|
||
|
||
def attach_features(ctx, acc_pos: np.ndarray, trades: List[Trade]) -> None:
|
||
bt_cfg = ctx.bt_cfg
|
||
confirms = [(ctx.an[tf], ctx.cuts[tf]) for tf in bt_cfg.confirmation_tfs]
|
||
# f_h1_ext_atr: the H1 frame wherever it sits; f_c_other_ext_atr: the first
|
||
# non-H1 confirmation with >= 21 bars (evaluated per bar on the device).
|
||
if bt_cfg.primary_tf == Timeframe.H1:
|
||
h1 = (1, None, None)
|
||
elif Timeframe.H1 in ctx.dfs_full and Timeframe.H1 in ctx.cuts:
|
||
h1 = (2, ctx.dev[Timeframe.H1], ctx.cuts[Timeframe.H1])
|
||
else:
|
||
h1 = (0, None, None)
|
||
others = [(ctx.dev[tf], ctx.cuts[tf]) for tf in bt_cfg.confirmation_tfs if tf is not Timeframe.H1][:2]
|
||
F = ck.features(acc_pos, ctx.dev[bt_cfg.primary_tf], ctx.an[bt_cfg.primary_tf], ctx.k2,
|
||
len(bt_cfg.confirmation_tfs), confirms, h1, others, ctx.flags, ctx.pf, ctx.pi)
|
||
for t, row in zip(trades, F):
|
||
t.features = {name: float(row[k]) for k, name in enumerate(position_sizing.FEATURE_ORDER)}
|
||
|
||
|
||
def _tf_block(row: K1Row, tf_name: str, role: str, direction: str, two: bool) -> Dict:
|
||
inds = describe_indicators(row, two)
|
||
is_buy = direction == "BUY"
|
||
out_inds: Dict[str, Dict] = {}
|
||
for name, ind in inds.items():
|
||
sig = ind["signal"]
|
||
base = float(ind["weight"])
|
||
eff = replay_record._finite(ind["effective_weight"])
|
||
out_inds[name] = {
|
||
"signal": sig,
|
||
"signal_name": ck.SIGNAL_NAMES[sig],
|
||
"value": replay_record._finite(ind["value"]),
|
||
"confidence": replay_record._finite(ind["confidence"]),
|
||
"base_weight": base,
|
||
"regime_mult": (eff / base) if (eff is not None and base) and base > 0 else None,
|
||
"eff_weight": eff,
|
||
"wscore": replay_record._finite(ind["weighted_score"]),
|
||
"agrees": replay_record._agrees(sig, direction),
|
||
"description": str(ind["description"] or ""),
|
||
}
|
||
kl = key_levels(row, two)
|
||
return {
|
||
"tf": tf_name,
|
||
"role": role,
|
||
"regime": ck.REGIME_NAMES[row.i["regime"]],
|
||
"vol": ck.VOL_NAMES[row.i["vol"]],
|
||
"composite_score": replay_record._finite(row.f["composite_score"]),
|
||
"composite_signal": int(row.i["composite_signal"]),
|
||
"confidence": replay_record._finite(row.f["composite_conf"]),
|
||
"atr": replay_record._finite(kl.get("atr")),
|
||
"atr_pct": replay_record._finite(kl.get("atr_pct")),
|
||
"pattern_signals": pattern_signals(row),
|
||
"indicators": out_inds,
|
||
}
|
||
|
||
|
||
def attach_snapshots(ctx, acc_pos: np.ndarray, trades: List[Trade]) -> None:
|
||
"""replay_record.snapshot_report for every accepted trade, from the device outputs."""
|
||
bt_cfg, platform_cfg = ctx.bt_cfg, ctx.platform_cfg
|
||
primary_df = ctx.dfs_full[bt_cfg.primary_tf]
|
||
primary_dur = pd.Timedelta(seconds=BAR_SECONDS[bt_cfg.primary_tf])
|
||
two = platform_cfg.technical.ema_mode == "2ema"
|
||
pdev = ctx.dev[bt_cfg.primary_tf]
|
||
medians = ck.prefix_median(pdev["high"], pdev["low"], pdev["n"])[acc_pos].get() if len(acc_pos) else []
|
||
ints, floats, wc, wa = _k2_rows(ctx.k2, acc_pos)
|
||
prows = k1_rows(ctx.an[bt_cfg.primary_tf], acc_pos)
|
||
conf_rows = {}
|
||
conf_cuts = {}
|
||
for tf in bt_cfg.confirmation_tfs:
|
||
cut = ctx.cuts[tf][acc_pos].get() if len(acc_pos) else np.zeros(0, np.int64)
|
||
conf_cuts[tf] = cut
|
||
present = cut > 0
|
||
rows = k1_rows(ctx.an[tf], np.where(present, cut - 1, 0))
|
||
conf_rows[tf] = rows
|
||
s = platform_cfg.signals
|
||
for m, (pos, t) in enumerate(zip(acc_pos, trades)):
|
||
pos = int(pos)
|
||
r2i, r2f, prow = ints[m], floats[m], prows[m]
|
||
r2f["_atr"] = prow.f["atr"] if prow.i["atr_present"] else 0.0
|
||
direction = t.direction
|
||
inds = describe_indicators(prow, two)
|
||
bullish, bearish = build_reasons(ctx, r2i, r2f, inds)
|
||
warnings, cleaned = build_warnings(ctx, pos, r2i, r2f, wc[m], wa[m])
|
||
tag = t.strategy_tag
|
||
pmask = r2i["penalty_mask"]
|
||
penalty_tags = {tagname: float(ck.PENALTY_AMOUNT[tagname]) for tagname in ck.PENALTY_SORTED
|
||
if pmask & (1 << ck.PENALTY_TAGS.index(tagname))}
|
||
mtf_signals = {bt_cfg.primary_tf.value: int(prow.i["composite_signal"])}
|
||
mtf_scores = {bt_cfg.primary_tf.value: replay_record._finite(prow.f["composite_score"])}
|
||
tfs = [_tf_block(prow, bt_cfg.primary_tf.value, "primary", direction, two)]
|
||
for k, tf in enumerate(bt_cfg.confirmation_tfs):
|
||
if conf_cuts[tf][m] == 0:
|
||
continue
|
||
crow = conf_rows[tf][m]
|
||
mtf_signals[tf.value] = int(crow.i["composite_signal"])
|
||
mtf_scores[tf.value] = replay_record._finite(crow.f["composite_score"])
|
||
if k + 1 < len(replay_record.ROLES):
|
||
tfs.append(_tf_block(crow, tf.value, replay_record.ROLES[k + 1], direction, two))
|
||
trail_arm = s.pb_trail_arm_r if tag == "XAU_PB" else s.trail_arm_r
|
||
trail_dist = s.pb_trail_dist_r if tag == "XAU_PB" else s.trail_dist_r
|
||
t.snapshot = {
|
||
"bar_ts": str(primary_df.index[pos]),
|
||
"decision_time": pd.Timestamp(primary_df.index[pos] + primary_dur).isoformat(),
|
||
"final_signal": int(r2i["final_signal"]),
|
||
"final_score": replay_record._finite(r2f["final_score"]),
|
||
"final_confidence": replay_record._finite(r2f["final_conf"]),
|
||
"quality": ck.QUALITY_NAMES[r2i["quality"]],
|
||
"mtf_agreement": bool(r2i["mtf_agreement"]),
|
||
"mtf_signals": mtf_signals,
|
||
"mtf_scores": mtf_scores,
|
||
"penalty_tags": penalty_tags,
|
||
"penalty_total": replay_record._finite(r2f["penalty_total"]),
|
||
"momentum_fading": bool(r2i["momentum_fading"]),
|
||
"exhaustion_count": int(r2i["exhaustion_count"]),
|
||
"spread_estimate": replay_record._finite(float(medians[m])),
|
||
"current_price": replay_record._finite(float(primary_df["Close"].iloc[pos])),
|
||
"warnings": warnings,
|
||
"cleaned_warnings": cleaned,
|
||
"bullish_reasons": bullish,
|
||
"bearish_reasons": bearish,
|
||
"setup": {
|
||
"strategy": tag,
|
||
"setup_pattern": ck.PATTERN_ORDER[r2i["cp_pattern"]] if tag == "XAU_CP" else None,
|
||
"round_trip_cost": replay_record._finite(s.round_trip_cost),
|
||
"trail_arm_r": replay_record._finite(trail_arm),
|
||
"trail_dist_r": replay_record._finite(trail_dist),
|
||
"entry_price": replay_record._finite(r2f["setup_entry"]),
|
||
"stop_loss": replay_record._finite(r2f["setup_sl"]),
|
||
"take_profit_1": replay_record._finite(r2f["setup_tp1"]),
|
||
"take_profit_2": replay_record._finite(r2f["setup_tp2"]),
|
||
"risk_reward": replay_record._finite(r2f["rr_rounded"]),
|
||
},
|
||
"tfs": tfs,
|
||
}
|
||
|
||
|
||
def build_bar_chunks(ctx):
|
||
"""replay_record's (matrix float32 [n,160], ts_ns, pos, warning0) for
|
||
every analysed primary bar (the K6 rows plus the host warning text)."""
|
||
bt_cfg = ctx.bt_cfg
|
||
k2 = ctx.k2
|
||
n_bars = k2.n_bars
|
||
if n_bars <= 0:
|
||
return None
|
||
status = k2.i("status").get()[:n_bars]
|
||
bar_pos = (np.flatnonzero(status >= 1) + k2.first_pos).astype(np.int64)
|
||
if len(bar_pos) == 0:
|
||
return None
|
||
confirms = [(ctx.an[tf], ctx.cuts[tf]) for tf in bt_cfg.confirmation_tfs]
|
||
mat = ck.bar_rows(bar_pos, ctx.dev[bt_cfg.primary_tf], ctx.an[bt_cfg.primary_tf], k2,
|
||
len(bt_cfg.confirmation_tfs), confirms)
|
||
ints, floats, wc, wa = _k2_rows(k2, bar_pos)
|
||
prows_atr = None
|
||
warning0: List[str] = []
|
||
prim = ctx.an[bt_cfg.primary_tf]
|
||
atr_present = prim.i("atr_present")[bar_pos].get()
|
||
atr_vals = prim.f("atr")[bar_pos].get()
|
||
for m, pos in enumerate(bar_pos):
|
||
r2i, r2f = ints[m], floats[m]
|
||
if r2i["warn_n"] == 0:
|
||
warning0.append("")
|
||
continue
|
||
r2f["_atr"] = float(atr_vals[m]) if atr_present[m] else 0.0
|
||
warnings, _ = build_warnings(ctx, int(pos), r2i, r2f, wc[m], wa[m])
|
||
warning0.append(warnings[0] if warnings else "")
|
||
primary_index = ctx.dfs_full[bt_cfg.primary_tf].index
|
||
return [(mat, primary_index.as_unit("ns").asi8[bar_pos].astype(np.int64), bar_pos.astype(np.int64), warning0)]
|
||
|
||
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
# CLI (the reference script's parser, config block and output, verbatim)
|
||
# ═════════════════════════════════════════════════════════════════════
|
||
|
||
def require_gpu_or_exit(requested_backend: Optional[str]) -> None:
|
||
if requested_backend == "cpu" or BACKEND.mode == "cpu":
|
||
raise SystemExit(
|
||
"cudaBasedBacktesting.py runs the decision stage on CUDA and has no CPU path "
|
||
"(--backend cpu / FB2_BACKEND=cpu requested). Use backtest/gpu_backtest.py for the CPU twins.")
|
||
if requested_backend:
|
||
BACKEND.set_mode(requested_backend)
|
||
if not BACKEND.available:
|
||
raise SystemExit(
|
||
f"cudaBasedBacktesting.py needs a working CUDA backend: {BACKEND.describe()}. "
|
||
f"Use backtest/gpu_backtest.py.")
|
||
|
||
|
||
def main(argv: Optional[List[str]] = None) -> int:
|
||
args = _parse_args(argv)
|
||
logging.basicConfig(
|
||
level=args.log_level.upper(),
|
||
format="%(asctime)s [%(levelname)s] %(name)s: %(message)s",
|
||
)
|
||
require_gpu_or_exit(args.backend)
|
||
|
||
pair = _resolve_pair(args.pair)
|
||
primary_tf = _resolve_tf(args.primary_tf)
|
||
confirm_tfs = (
|
||
[_resolve_tf(s.strip()) for s in args.confirm_tfs.split(",") if s.strip()]
|
||
if args.confirm_tfs is not None else None
|
||
)
|
||
from pair_signals import PAIR_PRODUCERS # noqa: E402
|
||
factory = PAIR_PRODUCERS.get(pair)
|
||
if factory is not None:
|
||
platform_cfg = factory(primary_tf, confirm_tfs)
|
||
else:
|
||
platform_cfg = PlatformConfig(
|
||
symbol=pair,
|
||
display_name=PAIR_INFO[pair]["display"],
|
||
primary_timeframe=primary_tf,
|
||
confirmation_timeframes=confirm_tfs or [Timeframe.H1, Timeframe.H4],
|
||
)
|
||
from app_config import apply_pair_overrides # noqa: E402
|
||
apply_pair_overrides(platform_cfg)
|
||
if args.fixed_rr_tp is not None:
|
||
if args.fixed_rr_tp <= 0:
|
||
raise SystemExit("--fixed-rr-tp must be > 0")
|
||
if (args.trail_arm is not None or args.trail_dist is not None or
|
||
any(s.split("=", 1)[0].strip().startswith(("signals.trail", "signals.pb_trail"))
|
||
for s in (args.set or []))):
|
||
raise SystemExit(
|
||
"--fixed-rr-tp conflicts with --trail-arm/--trail-dist/"
|
||
"--set signals.*trail* (the flag disables trailing entirely)")
|
||
if args.swing_sr_lookback > 0:
|
||
platform_cfg.technical.swing_sr_lookback = args.swing_sr_lookback
|
||
if args.trail_arm is not None:
|
||
platform_cfg.signals.trail_arm_r = args.trail_arm
|
||
if args.trail_dist is not None:
|
||
platform_cfg.signals.trail_dist_r = args.trail_dist
|
||
if args.cp:
|
||
platform_cfg.xau_strategy.cp_enabled = True
|
||
if args.cp_patterns:
|
||
platform_cfg.xau_strategy.cp_patterns = [
|
||
s.strip() for s in args.cp_patterns.split(",") if s.strip()
|
||
]
|
||
if args.big_candle is not None:
|
||
platform_cfg.xau_strategy.big_candle_atr_multiple = args.big_candle
|
||
if args.big_candle_hours is not None:
|
||
platform_cfg.xau_strategy.big_candle_lookback_hours = args.big_candle_hours
|
||
if args.set:
|
||
_apply_config_overrides(platform_cfg, args.set)
|
||
confirm_tfs = list(platform_cfg.confirmation_timeframes)
|
||
|
||
from datetime import datetime, timedelta, timezone
|
||
end = (
|
||
datetime.fromisoformat(args.end).replace(tzinfo=timezone.utc)
|
||
if args.end else datetime.now(timezone.utc)
|
||
)
|
||
if args.start:
|
||
start = datetime.fromisoformat(args.start).replace(tzinfo=timezone.utc)
|
||
else:
|
||
start = end - timedelta(days=args.days)
|
||
spread = (args.spread if args.spread is not None
|
||
else PAIR_INFO[pair].get("default_spread", 0.30))
|
||
bt_cfg = BacktestConfig(
|
||
pair=pair, start=start, end=end, primary_tf=primary_tf, confirmation_tfs=confirm_tfs,
|
||
warmup_bars=args.warmup_bars, cooldown_bars=args.cooldown_bars, max_hold_bars=args.max_hold_bars,
|
||
use_cache=not args.no_cache,
|
||
quality_filter=[q.strip() for q in args.quality.split(",") if q.strip()],
|
||
strategies_only=[("XAU_ATR" if s.strip() == "XAU" else s.strip())
|
||
for s in args.strategies.split(",") if s.strip()],
|
||
spread=spread, workers=args.workers, fixed_rr_tp=args.fixed_rr_tp or 0.0,
|
||
export_features=args.export_features, record_dir=args.record, record_bars=args.record_bars,
|
||
)
|
||
if args.workers:
|
||
logger.info(f"--workers {args.workers} is ignored: the replay runs on the device")
|
||
if args.record:
|
||
logger.info(
|
||
f"--record ON: decision snapshots → {args.record} "
|
||
f"(trades/indicators/events{'/bars' if args.record_bars else ''} parquet + manifest.json; "
|
||
f"the trade book is unchanged)")
|
||
if args.export_features:
|
||
logger.info(
|
||
f"--export-features ON: appending stake_multiplier, r_sized and "
|
||
f"{len(position_sizing.FEATURE_ORDER)} f_* feature columns to the "
|
||
f"CSV export (research flag; book itself is unchanged)")
|
||
logger.info(
|
||
f"Backend: {BACKEND.describe()} | "
|
||
f"Pair: {PAIR_INFO[pair]['display']} | Window: {start.date()} → {end.date()} | "
|
||
f"Confirm TFs: {','.join(tf.value for tf in confirm_tfs)} | Spread: {spread}"
|
||
)
|
||
|
||
trades, _ = run_backtest_cuda(bt_cfg, platform_cfg)
|
||
metrics = compute_metrics(trades)
|
||
print_report(pair, bt_cfg, metrics, trades=trades, detail=args.detail)
|
||
print_trade_reasoning(trades, limit=args.reasoning_samples)
|
||
if args.export_csv:
|
||
df = trades_to_dataframe(trades, include_features=args.export_features)
|
||
df.to_csv(args.export_csv, index=False)
|
||
logger.info(f"Exported {len(df)} trades → {args.export_csv}")
|
||
return 0
|
||
|
||
|
||
if __name__ == "__main__":
|
||
sys.exit(main())
|