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
287
prod/diag_acb_inject.py
Executable file
287
prod/diag_acb_inject.py
Executable file
@@ -0,0 +1,287 @@
|
||||
"""
|
||||
ACB-inject diagnostic.
|
||||
Pre-loads ACB factors via os.scandir (fast on remote mount, avoids materializing 15k files).
|
||||
Injects _day_base_boost/_day_beta directly after begin_day().
|
||||
Actor-style loop, static vol_ok, gidx fix → T=2155 base.
|
||||
Measures ROI impact of ACB boost from ng6_data eigenvalues.
|
||||
"""
|
||||
import sys, math, os, pathlib
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
sys.path.insert(0, '/mnt/dolphinng5_predict')
|
||||
sys.path.insert(0, '/mnt/dolphinng5_predict/nautilus_dolphin')
|
||||
|
||||
print("Importing...", flush=True)
|
||||
from nautilus_dolphin.nautilus.proxy_boost_engine import create_boost_engine
|
||||
print("Import done.", flush=True)
|
||||
|
||||
PARQUET_DIR = pathlib.Path('/mnt/dolphinng5_predict/vbt_cache')
|
||||
# Primary: dolphin_training (full Jan coverage); fallback: ng6_data (Jan21, Jan30-31)
|
||||
EIGENVALUES_PATHS = [
|
||||
pathlib.Path('/mnt/ng6_data/eigenvalues'), # extf backfill output (backfill_klines_exf.py)
|
||||
pathlib.Path('/mnt/dolphin_training/data/eigenvalues'), # dolphin_training archive (Dec31-Jan12)
|
||||
pathlib.Path('/mnt/dolphinng6_data/eigenvalues'), # ng3 live share (Jan2-8, Jan21, Jan30-31, Mar+)
|
||||
]
|
||||
VOL_P60_INWINDOW = 0.00009868
|
||||
|
||||
# ACB config constants (from adaptive_circuit_breaker.py)
|
||||
BETA_HIGH = 0.8
|
||||
BETA_LOW = 0.2
|
||||
W750_THRESHOLD_PCT = 60
|
||||
FUNDING_VERY_BEARISH = -0.0001
|
||||
FUNDING_BEARISH = 0.0
|
||||
DVOL_EXTREME = 80
|
||||
DVOL_ELEVATED = 55
|
||||
FNG_EXTREME_FEAR = 25
|
||||
FNG_FEAR = 40
|
||||
TAKER_SELLING = 0.8
|
||||
TAKER_MILD_SELLING = 0.9
|
||||
|
||||
ENG_KWARGS = dict(
|
||||
max_hold_bars=120, min_irp_alignment=0.45, max_leverage=8.0,
|
||||
vel_div_threshold=-0.02, vel_div_extreme=-0.05,
|
||||
min_leverage=0.5, leverage_convexity=3.0,
|
||||
fraction=0.20, fixed_tp_pct=0.0095, stop_pct=1.0,
|
||||
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, 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,
|
||||
)
|
||||
|
||||
def make_engine(cap=25000.0):
|
||||
eng = create_boost_engine(mode='d_liq', initial_capital=cap, **ENG_KWARGS)
|
||||
eng.set_esoteric_hazard_multiplier(0.0)
|
||||
return eng
|
||||
|
||||
def fast_get_npz_files(date_path, n=10):
|
||||
"""Get first n NPZ indicator files using os.scandir (avoids full glob sort)."""
|
||||
files = []
|
||||
try:
|
||||
with os.scandir(date_path) as it:
|
||||
for entry in it:
|
||||
if entry.name.endswith('__Indicators.npz') and entry.name.startswith('scan_'):
|
||||
files.append(pathlib.Path(entry.path))
|
||||
if len(files) >= n:
|
||||
break
|
||||
except Exception:
|
||||
pass
|
||||
return files
|
||||
|
||||
def load_acb_factors(date_str):
|
||||
"""Load funding/dvol/fng/taker and w750 for a date from any available source."""
|
||||
# Try each eigenvalues path in priority order
|
||||
date_path = None
|
||||
for ep in EIGENVALUES_PATHS:
|
||||
candidate = ep / date_str
|
||||
if candidate.exists():
|
||||
files = fast_get_npz_files(candidate, n=1)
|
||||
if files:
|
||||
date_path = candidate
|
||||
break
|
||||
|
||||
if date_path is None:
|
||||
return None
|
||||
|
||||
files = fast_get_npz_files(date_path, n=10)
|
||||
if not files:
|
||||
return None
|
||||
|
||||
indicators = {'funding_btc': [], 'dvol_btc': [], 'fng': [], 'taker': [], 'lambda_vel_w750': []}
|
||||
|
||||
for f in files:
|
||||
try:
|
||||
data = np.load(f, allow_pickle=True)
|
||||
# External factors from api_indicators
|
||||
if 'api_names' in data and 'api_indicators' in data:
|
||||
names = list(data['api_names'])
|
||||
vals = data['api_indicators']
|
||||
succ = data['api_success'] if 'api_success' in data else np.ones(len(names), dtype=bool)
|
||||
for key in ['funding_btc', 'dvol_btc', 'fng', 'taker']:
|
||||
if key in names:
|
||||
idx = names.index(key)
|
||||
if succ[idx] and np.isfinite(vals[idx]):
|
||||
indicators[key].append(float(vals[idx]))
|
||||
# w750 from scan_global
|
||||
if 'scan_global_names' in data and 'scan_global' in data:
|
||||
gnames = list(data['scan_global_names'])
|
||||
gvals = data['scan_global']
|
||||
if 'lambda_vel_w750' in gnames:
|
||||
idx = gnames.index('lambda_vel_w750')
|
||||
if idx < len(gvals) and np.isfinite(gvals[idx]):
|
||||
indicators['lambda_vel_w750'].append(float(gvals[idx]))
|
||||
except Exception:
|
||||
continue
|
||||
|
||||
result = {
|
||||
'funding_btc': float(np.median(indicators['funding_btc'])) if indicators['funding_btc'] else 0.0,
|
||||
'dvol_btc': float(np.median(indicators['dvol_btc'])) if indicators['dvol_btc'] else 50.0,
|
||||
'fng': float(np.median(indicators['fng'])) if indicators['fng'] else 50.0,
|
||||
'taker': float(np.median(indicators['taker'])) if indicators['taker'] else 1.0,
|
||||
'w750_vel': float(np.median(indicators['lambda_vel_w750'])) if indicators['lambda_vel_w750'] else 0.0,
|
||||
'available': True,
|
||||
}
|
||||
return result
|
||||
|
||||
def compute_signals(f):
|
||||
"""Replicate ACB get_cut_for_date signal count from factors dict."""
|
||||
signals = 0
|
||||
if f['funding_btc'] <= FUNDING_VERY_BEARISH:
|
||||
signals += 2
|
||||
elif f['funding_btc'] <= FUNDING_BEARISH:
|
||||
signals += 1
|
||||
dvol = f['dvol_btc']
|
||||
if dvol >= DVOL_EXTREME:
|
||||
signals += 2
|
||||
elif dvol >= DVOL_ELEVATED:
|
||||
signals += 1
|
||||
fng = f['fng']
|
||||
if fng <= FNG_EXTREME_FEAR:
|
||||
signals += 2
|
||||
elif fng <= FNG_FEAR:
|
||||
signals += 1
|
||||
taker = f['taker']
|
||||
if taker <= TAKER_SELLING:
|
||||
signals += 2
|
||||
elif taker <= TAKER_MILD_SELLING:
|
||||
signals += 1
|
||||
return signals
|
||||
|
||||
def compute_boost(signals):
|
||||
if signals >= 1.0:
|
||||
return 1.0 + 0.5 * math.log1p(signals)
|
||||
return 1.0
|
||||
|
||||
def preload_acb_all(all_dates):
|
||||
"""Load ACB factors for all dates. Returns {date_str: {boost, beta, ...}}."""
|
||||
print("Pre-loading ACB factors (fast scandir)...", flush=True)
|
||||
factors_by_date = {}
|
||||
w750_vals = []
|
||||
|
||||
for ds in all_dates:
|
||||
f = load_acb_factors(ds)
|
||||
if f:
|
||||
factors_by_date[ds] = f
|
||||
if f['w750_vel'] != 0.0:
|
||||
w750_vals.append(f['w750_vel'])
|
||||
|
||||
n_loaded = len(factors_by_date)
|
||||
w750_thresh = float(np.percentile(w750_vals, W750_THRESHOLD_PCT)) if w750_vals else 0.0
|
||||
print(f" Loaded: {n_loaded}/{len(all_dates)} dates, w750_thresh={w750_thresh:.6f}", flush=True)
|
||||
|
||||
# Compute boost + beta per date
|
||||
acb_by_date = {}
|
||||
for ds in all_dates:
|
||||
if ds in factors_by_date:
|
||||
f = factors_by_date[ds]
|
||||
signals = compute_signals(f)
|
||||
boost = compute_boost(signals)
|
||||
w750 = f['w750_vel']
|
||||
beta = BETA_HIGH if (w750_thresh == 0.0 or w750 >= w750_thresh) else BETA_LOW
|
||||
acb_by_date[ds] = {'boost': boost, 'beta': beta, 'signals': signals, 'w750': w750, 'available': True}
|
||||
if boost > 1.0 or beta != BETA_LOW:
|
||||
print(f" {ds}: signals={signals} boost={boost:.4f} beta={beta} w750={w750:.6f}", flush=True)
|
||||
else:
|
||||
# No data: boost=1.0, beta=midpoint (0.5) — unknown regime
|
||||
acb_by_date[ds] = {'boost': 1.0, 'beta': 0.5, 'signals': 0, 'w750': 0.0, 'available': False}
|
||||
|
||||
return acb_by_date
|
||||
|
||||
def compute_vol_ok(df):
|
||||
btc_f = df['BTCUSDT'].values.astype('float64')
|
||||
n = len(btc_f)
|
||||
vol_ok = np.zeros(n, dtype=bool)
|
||||
for j in range(50, n):
|
||||
seg = btc_f[max(0, j-50):j]
|
||||
diffs = np.diff(seg)
|
||||
denom = seg[:-1]
|
||||
if np.any(denom == 0):
|
||||
continue
|
||||
v = float(np.std(diffs / denom))
|
||||
if math.isfinite(v) and v > 0:
|
||||
vol_ok[j] = v > VOL_P60_INWINDOW
|
||||
return vol_ok
|
||||
|
||||
def run_day(df, date_str, eng, acb_info=None, nan_fix=True):
|
||||
eng.begin_day(date_str)
|
||||
# Inject ACB boost/beta directly after begin_day
|
||||
if acb_info:
|
||||
eng._day_base_boost = acb_info['boost']
|
||||
eng._day_beta = acb_info['beta']
|
||||
|
||||
data_arr = df.values
|
||||
cols = df.columns.tolist()
|
||||
vd_idx = cols.index('vel_div') if 'vel_div' in cols else -1
|
||||
v50_idx = cols.index('v50_lambda_max_velocity') if 'v50_lambda_max_velocity' in cols else -1
|
||||
v750_idx = cols.index('v750_lambda_max_velocity') if 'v750_lambda_max_velocity' in cols else -1
|
||||
i50_idx = cols.index('instability_50') if 'instability_50' in cols else -1
|
||||
usdt_idxs = [(c, cols.index(c)) for c in cols if c.endswith('USDT')]
|
||||
vol_ok = compute_vol_ok(df)
|
||||
trades = 0
|
||||
|
||||
for i in range(len(df)):
|
||||
row_vals = data_arr[i]
|
||||
vd_raw = float(row_vals[vd_idx]) if vd_idx != -1 else float('nan')
|
||||
if not math.isfinite(vd_raw):
|
||||
if nan_fix:
|
||||
eng._global_bar_idx += 1
|
||||
continue
|
||||
|
||||
v750 = float(row_vals[v750_idx]) if v750_idx != -1 and math.isfinite(float(row_vals[v750_idx])) else 0.0
|
||||
inst50 = float(row_vals[i50_idx]) if i50_idx != -1 and math.isfinite(float(row_vals[i50_idx])) else 0.0
|
||||
v50 = float(row_vals[v50_idx]) if v50_idx != -1 and math.isfinite(float(row_vals[v50_idx])) else 0.0
|
||||
prices = {sym: float(row_vals[ci]) for sym, ci in usdt_idxs
|
||||
if math.isfinite(float(row_vals[ci])) and float(row_vals[ci]) > 0}
|
||||
|
||||
prev_pos = eng.position
|
||||
if hasattr(eng, 'pre_bar_proxy_update'):
|
||||
eng.pre_bar_proxy_update(inst50, v750)
|
||||
|
||||
eng.step_bar(
|
||||
bar_idx=i, vel_div=vd_raw, prices=prices,
|
||||
v50_vel=v50, v750_vel=v750,
|
||||
vol_regime_ok=bool(vol_ok[i]),
|
||||
)
|
||||
if prev_pos is not None and eng.position is None:
|
||||
trades += 1
|
||||
|
||||
eng.end_day()
|
||||
return trades
|
||||
|
||||
def main():
|
||||
files = sorted(PARQUET_DIR.glob('*.parquet'))
|
||||
all_dates = [pf.stem for pf in files]
|
||||
print(f"Days: {len(files)}", flush=True)
|
||||
|
||||
acb_by_date = preload_acb_all(all_dates)
|
||||
|
||||
base_eng = make_engine()
|
||||
acb_eng = make_engine()
|
||||
base_T = acb_T = 0
|
||||
|
||||
have_eigen = set(d for d in all_dates if acb_by_date[d].get('available', False))
|
||||
|
||||
for pf in files:
|
||||
date_str = pf.stem
|
||||
df = pd.read_parquet(pf)
|
||||
|
||||
tb = run_day(df, date_str, base_eng, acb_info=None, nan_fix=True)
|
||||
ta = run_day(df, date_str, acb_eng, acb_info=acb_by_date[date_str], nan_fix=True)
|
||||
base_T += tb
|
||||
acb_T += ta
|
||||
|
||||
flag = '*' if date_str in have_eigen else ' '
|
||||
acb_d = acb_by_date[date_str]
|
||||
print(f"{date_str}{flag}[b={acb_d['boost']:.3f} β={acb_d['beta']:.1f} s={acb_d['signals']}]: "
|
||||
f"BASE+{tb:3d}(cum={base_T:4d} ${base_eng.capital:8.0f}) "
|
||||
f"ACB+{ta:3d}(cum={acb_T:4d} ${acb_eng.capital:8.0f})", flush=True)
|
||||
|
||||
ic = 25000.0
|
||||
print(f"\nBASELINE: T={base_T}, cap=${base_eng.capital:.2f}, ROI={100*(base_eng.capital/ic-1):.2f}%", flush=True)
|
||||
print(f"ACB: T={acb_T}, cap=${acb_eng.capital:.2f}, ROI={100*(acb_eng.capital/ic-1):.2f}%", flush=True)
|
||||
print(f"\nGold target: T=2155, ROI=+189.48%", flush=True)
|
||||
|
||||
if __name__ == '__main__':
|
||||
main()
|
||||
Reference in New Issue
Block a user