initial: import DOLPHIN baseline 2026-04-21 from dolphinng5_predict working tree
Includes core prod + GREEN/BLUE subsystems: - prod/ (BLUE harness, configs, scripts, docs) - nautilus_dolphin/ (GREEN Nautilus-native impl + dvae/ preserved) - adaptive_exit/ (AEM engine + models/bucket_assignments.pkl) - Observability/ (EsoF advisor, TUI, dashboards) - external_factors/ (EsoF producer) - mc_forewarning_qlabs_fork/ (MC regime/envelope) Excludes runtime caches, logs, backups, and reproducible artifacts per .gitignore.
This commit is contained in:
hjnormey
348
nautilus_dolphin/test_pf_klines_2y_experiment.py
Executable file
348
nautilus_dolphin/test_pf_klines_2y_experiment.py
Executable file
@@ -0,0 +1,348 @@
|
||||
"""2-Year Klines Fractal Experiment — Eigen Multi-Scale Divergence at 1-Minute Timescale.
|
||||
|
||||
Hypothesis: the eigenvalue velocity divergence principle (vel_div = w50_vel - w150_vel)
|
||||
holds fractally at any timescale. At 1-minute cadence:
|
||||
- w50 = 50 one-minute bars ≈ 50 min clock time (vs ~4 min in live DOLPHIN)
|
||||
- w150 = 150 one-minute bars ≈ 2.5 hr clock time (vs ~12.5 min in live DOLPHIN)
|
||||
|
||||
Threshold adaptation (NOT normalization — different timescale, not different universe):
|
||||
klines vel_div distribution (30-day sample, Jan 2024):
|
||||
median=+0.016, std=0.467, range [-8.13, +9.52]
|
||||
p7 ≈ -0.50 (matches champion ~7% signal rate)
|
||||
p2 ≈ -1.00 (matches champion extreme-rate ~2%)
|
||||
Champion NG3 thresholds: vel_div_threshold=-0.02 (p~7%), vel_div_extreme=-0.05
|
||||
Adapted klines thresholds: vel_div_threshold=-0.50, vel_div_extreme=-1.25
|
||||
|
||||
Data: vbt_cache_klines/ 2024-01-01 to 2026-03-05 (~795 days, 1-min klines → ARB512 eigenvalues)
|
||||
Asset universe: ~50 symbols (2024 = pre-STXUSDT era, NKNUSDT present)
|
||||
Note: universe shift mid-experiment expected — handled by ARS at daily level.
|
||||
|
||||
Full engine stack unchanged: ACBv6 + OB 4D (MockOB) + MC-Forewarner + EsoF(neutral) + ExF(neutral fallback)
|
||||
"""
|
||||
import sys, time, math, json, csv
|
||||
sys.stdout.reconfigure(encoding='utf-8', errors='replace')
|
||||
from pathlib import Path
|
||||
from datetime import datetime
|
||||
from collections import defaultdict
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).parent))
|
||||
|
||||
print("Compiling numba kernels...")
|
||||
t0c = time.time()
|
||||
from nautilus_dolphin.nautilus.alpha_asset_selector import compute_irp_nb, compute_ars_nb, rank_assets_irp_nb
|
||||
from nautilus_dolphin.nautilus.alpha_bet_sizer import compute_sizing_nb
|
||||
from nautilus_dolphin.nautilus.alpha_signal_generator import check_dc_nb
|
||||
from nautilus_dolphin.nautilus.ob_features import (
|
||||
OBFeatureEngine, compute_imbalance_nb, compute_depth_1pct_nb,
|
||||
compute_depth_quality_nb, compute_fill_probability_nb, compute_spread_proxy_nb,
|
||||
compute_depth_asymmetry_nb, compute_imbalance_persistence_nb,
|
||||
compute_withdrawal_velocity_nb, compute_market_agreement_nb, compute_cascade_signal_nb,
|
||||
)
|
||||
from nautilus_dolphin.nautilus.ob_provider import MockOBProvider
|
||||
|
||||
_p = np.array([1.0, 2.0, 3.0], dtype=np.float64)
|
||||
compute_irp_nb(_p, -1); compute_ars_nb(1.0, 0.5, 0.01)
|
||||
rank_assets_irp_nb(np.ones((10, 2), dtype=np.float64), 8, -1, 5, 500.0, 20, 0.20)
|
||||
compute_sizing_nb(-0.55, -0.50, -1.25, 3.0, 0.5, 5.0, 0.20, True, True, 0.0,
|
||||
np.zeros(4, dtype=np.int64), np.zeros(4, dtype=np.int64),
|
||||
np.zeros(5, dtype=np.float64), 0, -1, 0.01, 0.04)
|
||||
check_dc_nb(_p, 3, 1, 0.75)
|
||||
_b = np.array([100.0, 200.0, 300.0, 400.0, 500.0], dtype=np.float64)
|
||||
_a = np.array([110.0, 190.0, 310.0, 390.0, 510.0], dtype=np.float64)
|
||||
compute_imbalance_nb(_b, _a); compute_depth_1pct_nb(_b, _a)
|
||||
compute_depth_quality_nb(210.0, 200.0); compute_fill_probability_nb(1.0)
|
||||
compute_spread_proxy_nb(_b, _a); compute_depth_asymmetry_nb(_b, _a)
|
||||
compute_imbalance_persistence_nb(np.array([0.1, -0.1], dtype=np.float64), 2)
|
||||
compute_withdrawal_velocity_nb(np.array([100.0, 110.0], dtype=np.float64), 1)
|
||||
compute_market_agreement_nb(np.array([0.1, -0.05], dtype=np.float64), 2)
|
||||
compute_cascade_signal_nb(np.array([-0.05, -0.15], dtype=np.float64), 2, -0.10)
|
||||
print(f" JIT: {time.time() - t0c:.1f}s")
|
||||
|
||||
from nautilus_dolphin.nautilus.esf_alpha_orchestrator import NDAlphaEngine
|
||||
from nautilus_dolphin.nautilus.adaptive_circuit_breaker import AdaptiveCircuitBreaker
|
||||
from mc.mc_ml import DolphinForewarner
|
||||
|
||||
# ── Config ─────────────────────────────────────────────────────────────────────
|
||||
# klines cache: 2024-01-01 to 2026-03-05 stored in vbt_cache_klines/
|
||||
# (vbt_cache_klines = pure klines-derived parquets, 1-min cadence, no live 5s NG5 data mixed in)
|
||||
VBT_DIR = Path(r"C:\Users\Lenovo\Documents\- DOLPHIN NG HD HCM TSF Predict\vbt_cache_klines")
|
||||
DATE_START = '2024-01-01'
|
||||
DATE_END = '2026-03-05'
|
||||
|
||||
META_COLS = {'timestamp', 'scan_number', 'v50_lambda_max_velocity', 'v150_lambda_max_velocity',
|
||||
'v300_lambda_max_velocity', 'v750_lambda_max_velocity', 'vel_div',
|
||||
'instability_50', 'instability_150'}
|
||||
|
||||
# Threshold adapted for 1-min timescale (see module docstring for derivation)
|
||||
VD_THRESHOLD = -0.50 # p~7 (champion: -0.02 at NG3 scale)
|
||||
VD_EXTREME = -1.25 # p~2 (champion: -0.05 at NG3 scale, same 2.5× ratio)
|
||||
|
||||
ENGINE_KWARGS = dict(
|
||||
initial_capital=25000.0,
|
||||
vel_div_threshold=VD_THRESHOLD,
|
||||
vel_div_extreme=VD_EXTREME,
|
||||
min_leverage=0.5, max_leverage=5.0, leverage_convexity=3.0,
|
||||
fraction=0.20, fixed_tp_pct=0.0099, stop_pct=1.0, max_hold_bars=120,
|
||||
use_direction_confirm=True, dc_lookback_bars=7, dc_min_magnitude_bps=0.75,
|
||||
dc_skip_contradicts=True, dc_leverage_boost=1.0, dc_leverage_reduce=0.5,
|
||||
use_asset_selection=True, min_irp_alignment=0.45,
|
||||
use_sp_fees=True, use_sp_slippage=True,
|
||||
sp_maker_entry_rate=0.62, sp_maker_exit_rate=0.50,
|
||||
use_ob_edge=True, ob_edge_bps=5.0, ob_confirm_rate=0.40,
|
||||
lookback=100, use_alpha_layers=True, use_dynamic_leverage=True, seed=42,
|
||||
)
|
||||
|
||||
MC_MODELS_DIR = str(Path(r"C:\Users\Lenovo\Documents\- DOLPHIN NG HD HCM TSF Predict\nautilus_dolphin\mc_results\models"))
|
||||
MC_BASE_CFG = {
|
||||
'trial_id': 0,
|
||||
# MC-Forewarner was trained on champion range (-0.02/-0.05). The klines threshold
|
||||
# adaptation (-0.50/-1.25) is a timescale rescaling, not a capital-risk change.
|
||||
# Pass champion thresholds so MC assesses leverage/fraction risk correctly.
|
||||
'vel_div_threshold': -0.02, 'vel_div_extreme': -0.05,
|
||||
'use_direction_confirm': True, 'dc_lookback_bars': 7,
|
||||
'dc_min_magnitude_bps': 0.75, 'dc_skip_contradicts': True,
|
||||
'dc_leverage_boost': 1.00, 'dc_leverage_reduce': 0.50,
|
||||
'vd_trend_lookback': 10, 'min_leverage': 0.50,
|
||||
'max_leverage': 5.00,
|
||||
'leverage_convexity': 3.00, 'fraction': 0.20,
|
||||
'use_alpha_layers': True, 'use_dynamic_leverage': True,
|
||||
'fixed_tp_pct': 0.0099, 'stop_pct': 1.00, 'max_hold_bars': 120,
|
||||
'use_sp_fees': True, 'use_sp_slippage': True,
|
||||
'sp_maker_entry_rate': 0.62, 'sp_maker_exit_rate': 0.50,
|
||||
'use_ob_edge': True, 'ob_edge_bps': 5.00, 'ob_confirm_rate': 0.40,
|
||||
'ob_imbalance_bias': -0.09, 'ob_depth_scale': 1.00,
|
||||
'use_asset_selection': True, 'min_irp_alignment': 0.45, 'lookback': 100,
|
||||
'acb_beta_high': 0.80, 'acb_beta_low': 0.20, 'acb_w750_threshold_pct': 60,
|
||||
}
|
||||
|
||||
print("\nLoading MC-Forewarner trained models...")
|
||||
forewarner = DolphinForewarner(models_dir=MC_MODELS_DIR)
|
||||
print(" MC-Forewarner ready")
|
||||
|
||||
# ── Load klines parquet files (2024-2025 only) ─────────────────────────────────
|
||||
parquet_files = sorted(
|
||||
p for p in VBT_DIR.glob("*.parquet")
|
||||
if 'catalog' not in str(p) and DATE_START <= p.stem <= DATE_END
|
||||
)
|
||||
print(f"\nKlines parquet files: {len(parquet_files)} dates ({parquet_files[0].stem} to {parquet_files[-1].stem})")
|
||||
|
||||
# ── ACB init ───────────────────────────────────────────────────────────────────
|
||||
print("\nInitializing ACB v6...")
|
||||
acb = AdaptiveCircuitBreaker()
|
||||
date_strings = [pf.stem for pf in parquet_files]
|
||||
acb.preload_w750(date_strings)
|
||||
print(f" w750 p60 threshold: {acb._w750_threshold:.6f}")
|
||||
print(f" Dates with w750 data: {sum(1 for v in acb._w750_vel_cache.values() if v != 0.0)}/{len(date_strings)}")
|
||||
|
||||
# ── Vol p60 calibration from first 5 dates ─────────────────────────────────────
|
||||
all_vols = []
|
||||
for pf in parquet_files[:5]:
|
||||
df = pd.read_parquet(pf)
|
||||
if 'BTCUSDT' not in df.columns: continue
|
||||
pr = df['BTCUSDT'].values
|
||||
for i in range(60, len(pr)):
|
||||
seg = pr[max(0,i-50):i]
|
||||
if len(seg) < 10: continue
|
||||
v = float(np.std(np.diff(seg)/seg[:-1]))
|
||||
if v > 0: all_vols.append(v)
|
||||
vol_p60 = float(np.percentile(all_vols, 60)) if all_vols else 1e-4
|
||||
print(f"\nVol p60 (klines calibration): {vol_p60:.6f}")
|
||||
|
||||
# ── Pre-load all parquet data ───────────────────────────────────────────────────
|
||||
print(f"\nPre-loading {len(parquet_files)} parquet files...")
|
||||
t_load = time.time()
|
||||
pq_data = {}
|
||||
for i, pf in enumerate(parquet_files):
|
||||
df = pd.read_parquet(pf)
|
||||
ac = [c for c in df.columns if c not in META_COLS]
|
||||
bp = df['BTCUSDT'].values if 'BTCUSDT' in df.columns else None
|
||||
dv = np.full(len(df), np.nan)
|
||||
if bp is not None:
|
||||
for j in range(50, len(bp)):
|
||||
seg = bp[max(0,j-50):j]
|
||||
if len(seg) < 10: continue
|
||||
dv[j] = float(np.std(np.diff(seg)/seg[:-1]))
|
||||
pq_data[pf.stem] = (df, ac, dv)
|
||||
if (i+1) % 100 == 0:
|
||||
print(f" Loaded {i+1}/{len(parquet_files)} dates...")
|
||||
print(f" Done in {time.time()-t_load:.1f}s")
|
||||
|
||||
# ── ACB w750 cache: populate from klines parquet v750 column ───────────────────
|
||||
# NG3 NPZ indicator files don't exist for 2024-2025 klines dates, so preload_w750
|
||||
# above returned all-zero. Override with parquet-derived klines w750 velocities.
|
||||
print("\nPopulating ACB w750 cache from klines v750_lambda_max_velocity...")
|
||||
for date_str, (df, _, _) in pq_data.items():
|
||||
if 'v750_lambda_max_velocity' in df.columns:
|
||||
v750_vals = df['v750_lambda_max_velocity'].dropna()
|
||||
if len(v750_vals) > 0:
|
||||
acb._w750_vel_cache[date_str] = float(v750_vals.median())
|
||||
# Recompute threshold from klines w750 distribution
|
||||
_w750_vals = [v for v in acb._w750_vel_cache.values() if v != 0.0]
|
||||
if _w750_vals:
|
||||
acb._w750_threshold = float(np.percentile(_w750_vals, acb.config.W750_THRESHOLD_PCT))
|
||||
print(f" w750 klines p60 threshold: {acb._w750_threshold:.6f}")
|
||||
print(f" Dates with klines w750 data: {len(_w750_vals)}/{len(date_strings)}")
|
||||
else:
|
||||
print(" WARNING: no klines w750 data found — ACB beta will be constant")
|
||||
|
||||
# ── OB engine ──────────────────────────────────────────────────────────────────
|
||||
OB_ASSETS = ["BTCUSDT", "ETHUSDT", "BNBUSDT", "SOLUSDT"]
|
||||
_mock_ob = MockOBProvider(
|
||||
imbalance_bias=-0.09, depth_scale=1.0, assets=OB_ASSETS,
|
||||
imbalance_biases={"BTCUSDT": -0.086, "ETHUSDT": -0.092,
|
||||
"BNBUSDT": +0.05, "SOLUSDT": +0.05},
|
||||
)
|
||||
ob_eng = OBFeatureEngine(_mock_ob)
|
||||
ob_eng.preload_date("mock", OB_ASSETS)
|
||||
|
||||
# ── Full stack ─────────────────────────────────────────────────────────────────
|
||||
print(f"\n=== 2Y KLINES EXPERIMENT: ACBv6 + OB 4D + MC-Forewarner + EsoF(neutral) ===")
|
||||
print(f" Period: {DATE_START} to {DATE_END} ({len(parquet_files)} days)")
|
||||
print(f" vel_div threshold: {VD_THRESHOLD} (klines-adapted) extreme: {VD_EXTREME}")
|
||||
print(f" Timescale: 1-min bars w50=50min w150=2.5h max_hold=120min")
|
||||
t0 = time.time()
|
||||
|
||||
engine = NDAlphaEngine(**ENGINE_KWARGS)
|
||||
engine.set_ob_engine(ob_eng)
|
||||
engine.set_acb(acb)
|
||||
engine.set_mc_forewarner(forewarner, MC_BASE_CFG)
|
||||
engine.set_esoteric_hazard_multiplier(0.0) # EsoF neutral
|
||||
|
||||
# ── Per-date loop ──────────────────────────────────────────────────────────────
|
||||
all_daily = []
|
||||
all_trades = []
|
||||
|
||||
for date_str in date_strings:
|
||||
if date_str not in pq_data:
|
||||
continue
|
||||
df, asset_cols, dvol_arr = pq_data[date_str]
|
||||
|
||||
# Vol regime: per-bar boolean array matching champion API
|
||||
vol_ok = np.where(np.isfinite(dvol_arr), dvol_arr > vol_p60, False)
|
||||
|
||||
result = engine.process_day(date_str, df, asset_cols, vol_regime_ok=vol_ok)
|
||||
|
||||
all_daily.append({
|
||||
'date': date_str,
|
||||
'pnl': result.get('pnl', 0.0),
|
||||
'capital': result.get('capital', ENGINE_KWARGS['initial_capital']),
|
||||
'trades': result.get('trades', 0),
|
||||
'boost': result.get('boost', 1.0),
|
||||
'beta': result.get('beta', 0.0),
|
||||
'mc_status': result.get('mc_status', 'GREEN'),
|
||||
})
|
||||
# trade_log not in process_day result — accumulate from engine.trade_history per day
|
||||
|
||||
if len(all_daily) % 50 == 0:
|
||||
recent = all_daily[-50:]
|
||||
cum_cap = all_daily[-1]['capital']
|
||||
roi = (cum_cap - ENGINE_KWARGS['initial_capital']) / ENGINE_KWARGS['initial_capital'] * 100
|
||||
ntrades = sum(r['trades'] for r in recent)
|
||||
print(f" [{date_str}] Day {len(all_daily)}/{len(date_strings)} | ROI={roi:+.1f}% | Last-50d trades={ntrades}")
|
||||
|
||||
t_elapsed = time.time() - t0
|
||||
|
||||
# Collect all trades from engine history
|
||||
all_trades = [
|
||||
{'pnl': t.pnl_absolute, 'pnl_pct': t.pnl_pct * 100,
|
||||
'asset': t.asset, 'bars_held': t.bars_held,
|
||||
'entry_bar': t.entry_bar, 'exit_bar': t.exit_bar,
|
||||
'exit_reason': t.exit_reason, 'leverage': t.leverage}
|
||||
for t in engine.trade_history
|
||||
]
|
||||
|
||||
# ── Summary stats ──────────────────────────────────────────────────────────────
|
||||
capitals = [r['capital'] for r in all_daily]
|
||||
pnls = [r['pnl'] for r in all_daily]
|
||||
n_trades = sum(r['trades'] for r in all_daily)
|
||||
|
||||
final_cap = capitals[-1] if capitals else ENGINE_KWARGS['initial_capital']
|
||||
roi = (final_cap - ENGINE_KWARGS['initial_capital']) / ENGINE_KWARGS['initial_capital'] * 100
|
||||
|
||||
# Drawdown
|
||||
peak = ENGINE_KWARGS['initial_capital']
|
||||
max_dd = 0.0
|
||||
for c in capitals:
|
||||
if c > peak: peak = c
|
||||
dd = (peak - c) / peak * 100
|
||||
if dd > max_dd: max_dd = dd
|
||||
|
||||
# Sharpe (daily PnL / std)
|
||||
pnl_arr = np.array(pnls)
|
||||
sharpe = (pnl_arr.mean() / pnl_arr.std() * np.sqrt(252)) if pnl_arr.std() > 0 else 0.0
|
||||
|
||||
# Win rate
|
||||
wins = [r for r in all_trades if r.get('pnl', 0) > 0] if all_trades else []
|
||||
losses = [r for r in all_trades if r.get('pnl', 0) <= 0] if all_trades else []
|
||||
wr = len(wins) / (len(wins) + len(losses)) * 100 if (wins or losses) else 0.0
|
||||
|
||||
# Profit factor
|
||||
gross_win = sum(t.get('pnl', 0) for t in wins)
|
||||
gross_loss = abs(sum(t.get('pnl', 0) for t in losses))
|
||||
pf = gross_win / gross_loss if gross_loss > 0 else float('inf')
|
||||
|
||||
# Half-year comparison
|
||||
h1_dates = [r for r in all_daily if r['date'] < '2025-01-01']
|
||||
h2_dates = [r for r in all_daily if r['date'] >= '2025-01-01']
|
||||
h1_roi = sum(r['pnl'] for r in h1_dates) / ENGINE_KWARGS['initial_capital'] * 100
|
||||
h2_roi = sum(r['pnl'] for r in h2_dates) / ENGINE_KWARGS['initial_capital'] * 100
|
||||
|
||||
print(f"\n{'='*65}")
|
||||
print(f" 2Y KLINES EXPERIMENT RESULTS ({DATE_START} to {DATE_END})")
|
||||
print(f"{'='*65}")
|
||||
print(f" ROI: {roi:+.2f}%")
|
||||
print(f" PF: {pf:.3f}")
|
||||
print(f" Max DD: {max_dd:.2f}%")
|
||||
print(f" Sharpe: {sharpe:.2f}")
|
||||
print(f" Win Rate: {wr:.1f}%")
|
||||
print(f" Trades: {n_trades:,} ({n_trades/len(date_strings):.1f}/day avg)")
|
||||
print(f" Days: {len(all_daily)}")
|
||||
print(f" H1 ROI (2024): {h1_roi:+.2f}%")
|
||||
print(f" H2 ROI (2025): {h2_roi:+.2f}%")
|
||||
print(f" H2/H1 ratio: {h2_roi/h1_roi:.2f}x" if h1_roi != 0 else " H2/H1: N/A")
|
||||
print(f" Runtime: {t_elapsed/60:.1f} min")
|
||||
print(f"{'='*65}")
|
||||
print(f"\n Champion (55d NG3): ROI=+44.89% PF=1.123 DD=14.95% Sharpe=2.50 WR=49.3%")
|
||||
print(f" Timescale: klines 1-min vs champion 5s (12× longer bars)")
|
||||
print(f" Threshold: {VD_THRESHOLD} (klines p~7) vs champion -0.02 (NG3 p~7)")
|
||||
|
||||
# ── Save results ───────────────────────────────────────────────────────────────
|
||||
ts = datetime.now().strftime('%Y%m%d_%H%M%S')
|
||||
run_dir = Path(__file__).parent / 'run_logs'
|
||||
run_dir.mkdir(exist_ok=True)
|
||||
|
||||
summary = {
|
||||
'experiment': '2y_klines_fractal',
|
||||
'date_range': f'{DATE_START}_to_{DATE_END}',
|
||||
'timescale': '1min_klines',
|
||||
'vel_div_threshold': VD_THRESHOLD,
|
||||
'vel_div_extreme': VD_EXTREME,
|
||||
'roi_pct': roi, 'pf': pf, 'max_dd_pct': max_dd,
|
||||
'sharpe': sharpe, 'win_rate_pct': wr,
|
||||
'n_trades': n_trades, 'n_days': len(all_daily),
|
||||
'trades_per_day': n_trades / len(all_daily) if all_daily else 0,
|
||||
'h1_roi_pct': h1_roi, 'h2_roi_pct': h2_roi,
|
||||
'engine_kwargs': ENGINE_KWARGS,
|
||||
'runtime_s': t_elapsed,
|
||||
'run_ts': ts,
|
||||
}
|
||||
summary_path = run_dir / f'klines_2y_{ts}.json'
|
||||
with open(summary_path, 'w') as f:
|
||||
json.dump(summary, f, indent=2)
|
||||
print(f"\n Summary: {summary_path}")
|
||||
|
||||
if all_daily:
|
||||
daily_path = run_dir / f'klines_2y_daily_{ts}.csv'
|
||||
with open(daily_path, 'w', newline='') as f:
|
||||
w = csv.DictWriter(f, fieldnames=all_daily[0].keys())
|
||||
w.writeheader(); w.writerows(all_daily)
|
||||
print(f" Daily: {daily_path}")
|
||||
|
||||
if all_trades:
|
||||
trades_path = run_dir / f'klines_2y_trades_{ts}.csv'
|
||||
with open(trades_path, 'w', newline='') as f:
|
||||
w = csv.DictWriter(f, fieldnames=all_trades[0].keys())
|
||||
w.writeheader(); w.writerows(all_trades)
|
||||
print(f" Trades: {trades_path}")
|
||||
Reference in New Issue
Block a user