01c19662cb
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.
355 lines
16 KiB
Python
Executable File
355 lines
16 KiB
Python
Executable File
"""Stochastic resonance + noise injection experiment.
|
||
|
||
Question: can controlled randomness at specific points in a near-threshold system
|
||
improve performance? Three mechanistic hypotheses tested:
|
||
|
||
H1 — STOCHASTIC RESONANCE on vel_div signal (col: 'vel_div')
|
||
Bars hovering just above threshold (-0.02) occasionally fire with noise.
|
||
SR predicts: optimal sigma ≈ mean(distance_to_threshold) for near-miss bars.
|
||
Sigmas: 0.001 / 0.003 / 0.005 / 0.010 (5% / 15% / 25% / 50% of |threshold|)
|
||
|
||
H2 — PRICE DITHER on asset execution prices
|
||
Avoid fill clustering at round price levels. Small multiplicative noise.
|
||
Sigmas: 0.0001 (1bp) / 0.0005 (5bp)
|
||
|
||
H3 — TP TARGET DITHER per-run
|
||
Sensitivity analysis: does TP=99bps sit at a local optimum?
|
||
Sigma: 0.0001 (±1bp 1-sigma band around 0.0099)
|
||
|
||
Stats: Mann-Whitney U vs baseline, Cohen's d, 95% bootstrap CI on ROI delta.
|
||
Results saved incrementally → run_logs/noise_exp_YYYYMMDD_HHMMSS.csv
|
||
"""
|
||
import sys, time, math, csv, os
|
||
sys.stdout.reconfigure(encoding='utf-8', errors='replace')
|
||
from pathlib import Path
|
||
from datetime import datetime
|
||
from scipy.stats import mannwhitneyu
|
||
import numpy as np
|
||
import pandas as pd
|
||
|
||
HCM = Path(r"C:\Users\Lenovo\Documents\- DOLPHIN NG HD HCM TSF Predict")
|
||
sys.path.insert(0, str(HCM / "nautilus_dolphin"))
|
||
|
||
VBT_DIR = HCM / "vbt_cache"
|
||
MC_MODELS_DIR= str(HCM / "nautilus_dolphin" / "mc_results" / "models")
|
||
LOG_DIR = HCM / "nautilus_dolphin" / "run_logs"
|
||
LOG_DIR.mkdir(exist_ok=True)
|
||
|
||
# ── experiment parameters ───────────────────────────────────────────────────
|
||
N_SEEDS = 25 # per (noise_type, sigma) — 25 gives ~80% power to detect 3% ROI shift
|
||
|
||
CONFIGS = [
|
||
# (label, noise_type, sigma)
|
||
("baseline", "none", 0.0),
|
||
("sr_5pct", "signal_sr", 0.001),
|
||
("sr_15pct", "signal_sr", 0.003),
|
||
("sr_25pct", "signal_sr", 0.005),
|
||
("sr_50pct", "signal_sr", 0.010),
|
||
("price_1bp", "price_dither", 0.0001),
|
||
("price_5bp", "price_dither", 0.0005),
|
||
("tp_1bp", "tp_dither", 0.0001),
|
||
]
|
||
BASELINE_LABEL = "baseline"
|
||
|
||
# ── engine config (exact champion) ──────────────────────────────────────────
|
||
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'}
|
||
|
||
ENGINE_KWARGS = dict(
|
||
initial_capital=25000.0, vel_div_threshold=-0.02, vel_div_extreme=-0.05,
|
||
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_BASE_CFG = {
|
||
'trial_id': 0, 'vel_div_threshold': -0.020, 'vel_div_extreme': -0.050,
|
||
'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,
|
||
}
|
||
|
||
# ── JIT warmup (one-time) ────────────────────────────────────────────────────
|
||
print("JIT warmup...", end='', flush=True)
|
||
t_jit = 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(-0.03, -0.02, -0.05, 3.0, 0.5, 5.0, 0.20, True, True, 0.0,
|
||
np.zeros(4, np.int64), np.zeros(4, np.int64),
|
||
np.zeros(5, np.float64), 0, -1, 0.01, 0.04)
|
||
check_dc_nb(_p, 3, 1, 0.75)
|
||
_b = np.array([100., 200., 300., 400., 500.], dtype=np.float64)
|
||
_a = np.array([110., 190., 310., 390., 510.], dtype=np.float64)
|
||
compute_imbalance_nb(_b, _a); compute_depth_1pct_nb(_b, _a)
|
||
compute_depth_quality_nb(210., 200.); 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., 110.], 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" {time.time()-t_jit:.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
|
||
|
||
# ── load shared infrastructure (one-time) ───────────────────────────────────
|
||
print("Loading MC-Forewarner...", end='', flush=True)
|
||
forewarner = DolphinForewarner(models_dir=MC_MODELS_DIR)
|
||
print(" OK")
|
||
|
||
parquet_files = sorted([p for p in VBT_DIR.glob("*.parquet") if 'catalog' not in str(p)])
|
||
date_strings = [pf.stem for pf in parquet_files]
|
||
|
||
print("Initializing ACB...", end='', flush=True)
|
||
acb = AdaptiveCircuitBreaker()
|
||
acb.preload_w750(date_strings)
|
||
print(f" OK (w750 p60={acb._w750_threshold:.6f})")
|
||
|
||
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)
|
||
|
||
# ── pre-load all parquet data ────────────────────────────────────────────────
|
||
print("Pre-loading parquet data...", end='', flush=True)
|
||
all_vols = []
|
||
for pf in parquet_files[:2]:
|
||
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:
|
||
v = float(np.std(np.diff(seg) / seg[:-1]))
|
||
if v > 0: all_vols.append(v)
|
||
vol_p60 = float(np.percentile(all_vols, 60))
|
||
|
||
pq_data = {}
|
||
for pf in 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 i in range(50, len(bp)):
|
||
seg = bp[max(0, i-50):i]
|
||
if len(seg) >= 10:
|
||
dv[i] = float(np.std(np.diff(seg) / seg[:-1]))
|
||
pq_data[pf.stem] = (df, ac, dv)
|
||
print(f" {len(pq_data)} days")
|
||
|
||
# also log near-threshold vel_div distribution (SR calibration info)
|
||
all_vd = []
|
||
for ds, (df, ac, dv) in pq_data.items():
|
||
if 'vel_div' in df.columns:
|
||
vd_vals = df['vel_div'].dropna().values
|
||
all_vd.extend(vd_vals[(vd_vals > -0.05) & (vd_vals < 0.0)])
|
||
all_vd = np.array(all_vd)
|
||
near_thresh = all_vd[(all_vd > -0.025) & (all_vd < -0.015)]
|
||
print(f" vel_div near-threshold (-0.025 to -0.015): N={len(near_thresh)}, "
|
||
f"mean={np.mean(near_thresh):+.5f}, σ={np.std(near_thresh):.5f}")
|
||
print(f" SR optimal sigma ≈ mean distance to threshold: "
|
||
f"{float(np.mean(np.abs(near_thresh - (-0.02)))):.4f}")
|
||
|
||
# ── engine runner ────────────────────────────────────────────────────────────
|
||
def run_engine(data_dict, engine_kw, vol_p60_val):
|
||
eng = NDAlphaEngine(**engine_kw)
|
||
eng.set_ob_engine(ob_eng)
|
||
eng.set_acb(acb)
|
||
eng.set_mc_forewarner(forewarner, MC_BASE_CFG)
|
||
eng.set_esoteric_hazard_multiplier(0.0)
|
||
|
||
dstats = []
|
||
for pf in parquet_files:
|
||
ds = pf.stem
|
||
df, acols, dvol = data_dict[ds]
|
||
vol_ok = np.where(np.isfinite(dvol), dvol > vol_p60_val, False)
|
||
stats = eng.process_day(ds, df, acols, vol_regime_ok=vol_ok)
|
||
dstats.append({**stats, 'cap': eng.capital})
|
||
|
||
tr = eng.trade_history
|
||
wins = [t for t in tr if t.pnl_absolute > 0]
|
||
loss = [t for t in tr if t.pnl_absolute <= 0]
|
||
gw = sum(t.pnl_absolute for t in wins) if wins else 0.0
|
||
gl = abs(sum(t.pnl_absolute for t in loss)) if loss else 0.0
|
||
roi = (eng.capital - 25000.0) / 25000.0 * 100
|
||
pff = gw / gl if gl > 0 else 999.0
|
||
dr = np.array([s['pnl'] / 25000.0 * 100 for s in dstats])
|
||
sh = float(np.mean(dr) / np.std(dr) * np.sqrt(365)) if np.std(dr) > 0 else 0.0
|
||
pk = 25000.0; mdd = 0.0
|
||
for s in dstats:
|
||
pk = max(pk, s['cap']); mdd = max(mdd, (pk - s['cap']) / pk * 100)
|
||
wr = len(wins) / len(tr) * 100 if tr else 0.0
|
||
return dict(roi=roi, pf=pff, dd=mdd, sharpe=sh, wr=wr,
|
||
trades=len(tr), capital=eng.capital)
|
||
|
||
# ── noise application ────────────────────────────────────────────────────────
|
||
def apply_noise(noise_type, sigma, seed_val):
|
||
"""Return (data_dict_noisy, engine_kw_noisy)."""
|
||
rng = np.random.default_rng(seed_val)
|
||
ekw = dict(ENGINE_KWARGS)
|
||
|
||
if noise_type == "none":
|
||
return pq_data, ekw
|
||
|
||
if noise_type == "tp_dither":
|
||
tp_noise = float(rng.normal(0, sigma))
|
||
ekw = dict(ENGINE_KWARGS)
|
||
ekw['fixed_tp_pct'] = max(0.003, ENGINE_KWARGS['fixed_tp_pct'] + tp_noise)
|
||
return pq_data, ekw # no data modification
|
||
|
||
# signal_sr or price_dither — need data copies
|
||
noisy = {}
|
||
for ds, (df, ac, dvol) in pq_data.items():
|
||
df2 = df.copy()
|
||
|
||
if noise_type == "signal_sr":
|
||
if 'vel_div' in df2.columns:
|
||
noise = rng.normal(0, sigma, len(df2)).astype(np.float32)
|
||
df2['vel_div'] = df2['vel_div'] + noise
|
||
|
||
elif noise_type == "price_dither":
|
||
for col in ac:
|
||
if col in df2.columns:
|
||
noise = rng.normal(0, sigma, len(df2))
|
||
df2[col] = df2[col] * (1.0 + noise)
|
||
# recompute dvol from dithered BTC prices
|
||
if 'BTCUSDT' in df2.columns:
|
||
bp = df2['BTCUSDT'].values
|
||
dv2 = np.full(len(df2), np.nan)
|
||
for i in range(50, len(bp)):
|
||
seg = bp[max(0, i-50):i]
|
||
if len(seg) >= 10:
|
||
dv2[i] = float(np.std(np.diff(seg) / seg[:-1]))
|
||
dvol = dv2
|
||
|
||
noisy[ds] = (df2, ac, dvol)
|
||
return noisy, ekw
|
||
|
||
# ── incremental CSV output ───────────────────────────────────────────────────
|
||
run_ts = datetime.now().strftime("%Y%m%d_%H%M%S")
|
||
out_path = LOG_DIR / f"noise_exp_{run_ts}.csv"
|
||
FIELDS = ['label', 'noise_type', 'sigma', 'seed',
|
||
'roi', 'pf', 'dd', 'sharpe', 'wr', 'trades', 'capital', 'elapsed_s']
|
||
|
||
with open(out_path, 'w', newline='') as f:
|
||
csv.writer(f).writerow(FIELDS)
|
||
|
||
def append_row(row_dict):
|
||
with open(out_path, 'a', newline='') as f:
|
||
csv.writer(f).writerow([row_dict[k] for k in FIELDS])
|
||
|
||
# ── main experiment loop ─────────────────────────────────────────────────────
|
||
print(f"\n{'='*65}")
|
||
print(f" NOISE EXPERIMENT — {len(CONFIGS)} configs × up to {N_SEEDS} seeds")
|
||
print(f" Output: {out_path.name}")
|
||
print(f"{'='*65}")
|
||
|
||
all_results = {} # label → list of roi values (for final stats)
|
||
t_exp_start = time.time()
|
||
completed = 0
|
||
total_runs = 1 + (len(CONFIGS) - 1) * N_SEEDS # baseline=1, others=N_SEEDS
|
||
|
||
for label, noise_type, sigma in CONFIGS:
|
||
n = 1 if noise_type == "none" else N_SEEDS
|
||
rois = []
|
||
print(f"\n [{label}] noise={noise_type} σ={sigma} n={n}")
|
||
|
||
for seed_i in range(n):
|
||
t0 = time.time()
|
||
data_d, eng_kw = apply_noise(noise_type, sigma, seed_val=seed_i + 1000)
|
||
result = run_engine(data_d, eng_kw, vol_p60)
|
||
elapsed = time.time() - t0
|
||
rois.append(result['roi'])
|
||
row = dict(label=label, noise_type=noise_type, sigma=sigma, seed=seed_i,
|
||
elapsed_s=round(elapsed, 1), **{k: round(result[k], 4) for k in result})
|
||
append_row(row)
|
||
completed += 1
|
||
eta_s = (time.time() - t_exp_start) / completed * (total_runs - completed)
|
||
print(f" seed={seed_i:2d} ROI={result['roi']:+6.2f}% PF={result['pf']:.3f}"
|
||
f" DD={result['dd']:.2f}% T={result['trades']}"
|
||
f" [{elapsed:.0f}s | ETA {eta_s/60:.0f}min]")
|
||
|
||
all_results[label] = rois
|
||
|
||
# ── final analysis ───────────────────────────────────────────────────────────
|
||
print(f"\n{'='*65}")
|
||
print(f" RESULTS SUMMARY")
|
||
print(f"{'='*65}")
|
||
baseline_rois = all_results.get(BASELINE_LABEL, [44.89])
|
||
b_roi = float(np.mean(baseline_rois))
|
||
print(f" {'Label':<14} {'σ':>8} {'E[ROI]':>8} {'±std':>7} {'ΔROI':>7} "
|
||
f"{'Cohen_d':>8} {'MW_p':>6} {'Beat%':>6} {'E[PF]':>6} {'E[T]':>6}")
|
||
print(f" {'-'*90}")
|
||
|
||
for label, noise_type, sigma in CONFIGS:
|
||
rois = all_results[label]
|
||
df_res = pd.read_csv(out_path)
|
||
sub = df_res[df_res['label'] == label]
|
||
mean_roi = float(np.mean(rois))
|
||
std_roi = float(np.std(rois)) if len(rois) > 1 else 0.0
|
||
delta = mean_roi - b_roi
|
||
mean_pf = float(sub['pf'].mean())
|
||
mean_t = float(sub['trades'].mean())
|
||
beat_pct = float(np.mean(np.array(rois) > b_roi)) * 100 if len(rois) > 1 else (100 if mean_roi > b_roi else 0)
|
||
|
||
# Cohen's d vs baseline
|
||
if len(rois) > 1 and len(baseline_rois) > 1:
|
||
pooled_std = math.sqrt((np.var(rois) + np.var(baseline_rois)) / 2)
|
||
cohens_d = delta / pooled_std if pooled_std > 0 else 0.0
|
||
else:
|
||
cohens_d = 0.0
|
||
|
||
# Mann-Whitney U vs baseline
|
||
if len(rois) > 1 and len(baseline_rois) > 1:
|
||
try:
|
||
_, mw_p = mannwhitneyu(rois, baseline_rois, alternative='two-sided')
|
||
except Exception:
|
||
mw_p = 1.0
|
||
else:
|
||
mw_p = 1.0
|
||
|
||
print(f" {label:<14} {sigma:>8.4f} {mean_roi:>+7.2f}% {std_roi:>6.2f}%"
|
||
f" {delta:>+6.2f}% {cohens_d:>+8.3f} {mw_p:>6.3f} {beat_pct:>5.1f}%"
|
||
f" {mean_pf:>6.3f} {mean_t:>6.0f}")
|
||
|
||
print(f"{'='*65}")
|
||
print(f"\n Interpretation guide:")
|
||
print(f" ΔROI > 0 & MW_p < 0.10 & Cohen_d > 0.3 → promising signal")
|
||
print(f" SR optimal sigma ≈ mean near-threshold distance (see calibration above)")
|
||
print(f"\n Full results → {out_path}")
|
||
print(f" Total time: {(time.time()-t_exp_start)/60:.1f} min")
|