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
309
nautilus_dolphin/exf_regime_characterizer.py
Executable file
309
nautilus_dolphin/exf_regime_characterizer.py
Executable file
@@ -0,0 +1,309 @@
|
||||
"""ExF-Driven Regime Characterizer
|
||||
==================================
|
||||
Reads the backfilled ExF NPZ files for all klines dates and builds a
|
||||
daily regime score from: funding_btc, dvol_btc, fng, taker.
|
||||
|
||||
Then cross-tabulates:
|
||||
1. ExF composite score vs vel_div SHORT/LONG edge
|
||||
2. Each ExF indicator individually vs edge
|
||||
3. Can ExF predict which DAYS/QUARTERS will have edge?
|
||||
|
||||
Also attempts to characterize "overall posture" (BEAR/NEUTRAL/BULL) from ExF
|
||||
and validate against the year-level edge pattern we observed:
|
||||
2021/22 (bear/vol) = high edge, 2023 (bull/low-vol) = near-zero edge.
|
||||
|
||||
Outputs:
|
||||
run_logs/exf_regime_YYYYMMDD.csv — per-date ExF values + edge
|
||||
run_logs/exf_regime_YYYYMMDD.json — summary correlations + regime bins
|
||||
Console: correlation table + regime breakdown
|
||||
"""
|
||||
import sys, time, json, csv, gc
|
||||
sys.stdout.reconfigure(encoding='utf-8', errors='replace')
|
||||
from pathlib import Path
|
||||
from datetime import datetime, date
|
||||
from collections import defaultdict
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
VBT_DIR = Path(r"C:\Users\Lenovo\Documents\- DOLPHIN NG HD HCM TSF Predict\vbt_cache_klines")
|
||||
EIGEN_PATH = Path(r"C:\Users\Lenovo\Documents\- Dolphin NG HD (NG3)\correlation_arb512\eigenvalues")
|
||||
LOG_DIR = Path(r"C:\Users\Lenovo\Documents\- DOLPHIN NG HD HCM TSF Predict\nautilus_dolphin\run_logs")
|
||||
TP_BPS = 95
|
||||
MAX_HOLD = 120
|
||||
tp_pct = TP_BPS / 10000.0
|
||||
SHORT_T = -0.020
|
||||
LONG_T = +0.020
|
||||
|
||||
# ExF indicator names as stored in NPZ files
|
||||
EXF_NAMES = ['funding_btc', 'dvol_btc', 'fng', 'taker']
|
||||
|
||||
def load_exf_for_date(date_str):
|
||||
"""Load ExF indicators for a date. Returns dict of name->value or None."""
|
||||
date_path = EIGEN_PATH / date_str
|
||||
if not date_path.exists():
|
||||
return None
|
||||
npz_files = sorted(date_path.glob('scan_*__Indicators.npz'))
|
||||
if not npz_files:
|
||||
return None
|
||||
try:
|
||||
data = np.load(str(npz_files[0]), allow_pickle=True)
|
||||
if 'api_names' not in data or 'api_indicators' not in data:
|
||||
return None
|
||||
names = list(data['api_names'])
|
||||
vals = data['api_indicators']
|
||||
result = {}
|
||||
for name in EXF_NAMES:
|
||||
if name in names:
|
||||
idx = names.index(name)
|
||||
v = float(vals[idx])
|
||||
result[name] = v if np.isfinite(v) else None
|
||||
else:
|
||||
result[name] = None
|
||||
return result
|
||||
except Exception:
|
||||
return None
|
||||
|
||||
parquet_files = sorted(VBT_DIR.glob("*.parquet"))
|
||||
parquet_files = [p for p in parquet_files if 'catalog' not in str(p)]
|
||||
total = len(parquet_files)
|
||||
print(f"Files: {total} Loading ExF + computing raw edge per date...")
|
||||
print(f"SHORT_T={SHORT_T} LONG_T={LONG_T} TP={TP_BPS}bps MAX_HOLD={MAX_HOLD}\n")
|
||||
|
||||
# Control
|
||||
ctrl_dn_count = ctrl_up_count = ctrl_n = 0
|
||||
|
||||
rows = []
|
||||
t0 = time.time()
|
||||
|
||||
for i, pf in enumerate(parquet_files):
|
||||
ds = pf.stem
|
||||
year = ds[:4]
|
||||
month = int(ds[5:7])
|
||||
quarter = f"Q{(month-1)//3+1}"
|
||||
|
||||
# Load ExF
|
||||
exf = load_exf_for_date(ds)
|
||||
funding = exf.get('funding_btc') if exf else None
|
||||
dvol = exf.get('dvol_btc') if exf else None
|
||||
fng = exf.get('fng') if exf else None
|
||||
taker = exf.get('taker') if exf else None
|
||||
|
||||
try:
|
||||
df = pd.read_parquet(pf)
|
||||
except Exception:
|
||||
continue
|
||||
if 'vel_div' not in df.columns or 'BTCUSDT' not in df.columns:
|
||||
continue
|
||||
|
||||
vd = df['vel_div'].values.astype(np.float64)
|
||||
btc = df['BTCUSDT'].values.astype(np.float64)
|
||||
vd = np.where(np.isfinite(vd), vd, 0.0)
|
||||
btc = np.where(np.isfinite(btc) & (btc > 0), btc, np.nan)
|
||||
n = len(btc)
|
||||
if n < MAX_HOLD + 5:
|
||||
del df, vd, btc
|
||||
continue
|
||||
|
||||
# BTC day metrics
|
||||
btc_clean = btc[np.isfinite(btc)]
|
||||
btc_return = (btc_clean[-1] - btc_clean[0]) / btc_clean[0] if len(btc_clean) >= 2 else 0.0
|
||||
log_r = np.diff(np.log(btc_clean)) if len(btc_clean) >= 2 else np.array([])
|
||||
realized_vol = float(np.std(log_r)) if len(log_r) > 1 else 0.0
|
||||
|
||||
# vel_div distribution
|
||||
vd_finite = vd[np.isfinite(vd)]
|
||||
vd_p10 = float(np.percentile(vd_finite, 10)) if len(vd_finite) > 5 else 0.0
|
||||
vd_p50 = float(np.median(vd_finite)) if len(vd_finite) > 0 else 0.0
|
||||
vd_p90 = float(np.percentile(vd_finite, 90)) if len(vd_finite) > 5 else 0.0
|
||||
|
||||
# Control
|
||||
for j in range(0, n - MAX_HOLD, 60):
|
||||
ep = btc[j]; ex = btc[j + MAX_HOLD]
|
||||
if np.isfinite(ep) and np.isfinite(ex) and ep > 0:
|
||||
r = (ex - ep) / ep
|
||||
ctrl_dn_count += int(r <= -tp_pct)
|
||||
ctrl_up_count += int(r >= tp_pct)
|
||||
ctrl_n += 1
|
||||
|
||||
# Raw edge per direction
|
||||
n_usable = n - MAX_HOLD
|
||||
from numpy.lib.stride_tricks import sliding_window_view
|
||||
windows = sliding_window_view(btc, MAX_HOLD + 1)[:n_usable]
|
||||
ep_arr = windows[:, 0]
|
||||
fut_min = np.nanmin(windows[:, 1:], axis=1)
|
||||
fut_max = np.nanmax(windows[:, 1:], axis=1)
|
||||
valid = np.isfinite(ep_arr) & (ep_arr > 0)
|
||||
|
||||
for direction, threshold in [('S', SHORT_T), ('L', LONG_T)]:
|
||||
active = (vd[:n_usable] <= threshold) if direction == 'S' else (vd[:n_usable] >= threshold)
|
||||
sig_idx = np.where(active & valid)[0]
|
||||
n_sig = len(sig_idx)
|
||||
if n_sig == 0:
|
||||
rows.append({
|
||||
'date': ds, 'year': year, 'quarter': quarter,
|
||||
'direction': direction,
|
||||
'n_sig': 0, 'wr': None, 'n_trades': 0,
|
||||
'funding_btc': funding, 'dvol_btc': dvol, 'fng': fng, 'taker': taker,
|
||||
'btc_return': round(btc_return, 6),
|
||||
'realized_vol': round(realized_vol, 8),
|
||||
'vd_p10': round(vd_p10, 6), 'vd_p50': round(vd_p50, 6), 'vd_p90': round(vd_p90, 6),
|
||||
})
|
||||
continue
|
||||
ep_s = ep_arr[sig_idx]
|
||||
if direction == 'S':
|
||||
hit = fut_min[sig_idx] <= ep_s * (1.0 - tp_pct)
|
||||
else:
|
||||
hit = fut_max[sig_idx] >= ep_s * (1.0 + tp_pct)
|
||||
wins = int(np.sum(hit))
|
||||
n_tr = len(sig_idx)
|
||||
wr = wins / n_tr * 100 if n_tr > 0 else None
|
||||
rows.append({
|
||||
'date': ds, 'year': year, 'quarter': quarter,
|
||||
'direction': direction,
|
||||
'n_sig': n_sig, 'wr': round(wr, 3) if wr is not None else None, 'n_trades': n_tr,
|
||||
'funding_btc': funding, 'dvol_btc': dvol, 'fng': fng, 'taker': taker,
|
||||
'btc_return': round(btc_return, 6),
|
||||
'realized_vol': round(realized_vol, 8),
|
||||
'vd_p10': round(vd_p10, 6), 'vd_p50': round(vd_p50, 6), 'vd_p90': round(vd_p90, 6),
|
||||
})
|
||||
|
||||
del df, vd, btc, windows, ep_arr, fut_min, fut_max, valid
|
||||
|
||||
if (i + 1) % 100 == 0:
|
||||
gc.collect()
|
||||
elapsed = time.time() - t0
|
||||
print(f" [{i+1}/{total}] {ds} {elapsed/60:.1f}m")
|
||||
|
||||
elapsed = time.time() - t0
|
||||
print(f"\nPass complete: {elapsed:.0f}s")
|
||||
|
||||
ctrl_dn_pct = ctrl_dn_count / ctrl_n * 100 if ctrl_n else 7.1
|
||||
ctrl_up_pct = ctrl_up_count / ctrl_n * 100 if ctrl_n else 7.1
|
||||
print(f"Control: DOWN={ctrl_dn_pct:.2f}% UP={ctrl_up_pct:.2f}% n={ctrl_n:,}")
|
||||
|
||||
# Analysis
|
||||
df_rows = pd.DataFrame(rows)
|
||||
|
||||
# Add edge column
|
||||
df_rows['baseline'] = df_rows['direction'].map({'S': ctrl_dn_pct, 'L': ctrl_up_pct})
|
||||
df_rows['edge'] = df_rows['wr'] - df_rows['baseline']
|
||||
|
||||
print(f"\n{'='*70}")
|
||||
print(f" ExF AVAILABILITY")
|
||||
print(f"{'='*70}")
|
||||
for col in ['funding_btc', 'dvol_btc', 'fng', 'taker']:
|
||||
n_avail = df_rows[df_rows['direction']=='S'][col].notna().sum()
|
||||
print(f" {col:<15}: {n_avail}/{total} dates available ({n_avail/total*100:.0f}%)")
|
||||
|
||||
# Correlations (SHORT only, dates with trades)
|
||||
df_s = df_rows[(df_rows['direction'] == 'S') & (df_rows['n_trades'] > 0)].copy()
|
||||
print(f"\n{'='*70}")
|
||||
print(f" CORRELATIONS WITH SHORT EDGE (n={len(df_s)} days with trades)")
|
||||
print(f"{'='*70}")
|
||||
for col in ['funding_btc', 'dvol_btc', 'fng', 'taker', 'btc_return', 'realized_vol', 'vd_p10', 'vd_p50']:
|
||||
sub = df_s[[col, 'edge']].dropna()
|
||||
if len(sub) < 10:
|
||||
print(f" {col:<18}: n={len(sub)} (insufficient)")
|
||||
continue
|
||||
corr = sub[col].corr(sub['edge'])
|
||||
print(f" {col:<18}: r={corr:>+.4f} (n={len(sub)})")
|
||||
|
||||
# ExF bin analysis
|
||||
print(f"\n{'='*70}")
|
||||
print(f" EDGE BY ExF QUARTILE (SHORT)")
|
||||
print(f"{'='*70}")
|
||||
for col in ['funding_btc', 'dvol_btc', 'fng', 'taker']:
|
||||
sub = df_s[[col, 'edge']].dropna()
|
||||
if len(sub) < 20:
|
||||
continue
|
||||
try:
|
||||
sub['q'] = pd.qcut(sub[col], 4, labels=False, duplicates='drop')
|
||||
sub['q_label'] = sub['q'].map(lambda x: ['Q1(low)', 'Q2', 'Q3', 'Q4(high)'][int(x)] if pd.notna(x) else 'NA')
|
||||
except Exception:
|
||||
sub['q'] = pd.cut(sub[col], 4, labels=False)
|
||||
sub['q_label'] = sub['q'].map(lambda x: f'B{int(x)}' if pd.notna(x) else 'NA')
|
||||
print(f"\n {col}:")
|
||||
for q_label, grp in sub.groupby('q_label', observed=True):
|
||||
print(f" {q_label:<12}: mean_edge={grp['edge'].mean():>+.2f}pp n={len(grp)}")
|
||||
|
||||
# Realized vol quartile
|
||||
print(f"\n{'='*70}")
|
||||
print(f" EDGE BY REALIZED_VOL QUARTILE (SHORT)")
|
||||
print(f"{'='*70}")
|
||||
sub = df_s[['realized_vol', 'edge']].dropna()
|
||||
sub['q'] = pd.qcut(sub['realized_vol'], 4, labels=['Q1(calm)', 'Q2', 'Q3', 'Q4(volatile)'])
|
||||
for q_label, grp in sub.groupby('q', observed=True):
|
||||
print(f" {q_label:<15}: mean_edge={grp['edge'].mean():>+.2f}pp median={grp['edge'].median():>+.2f}pp n={len(grp)}")
|
||||
|
||||
# BTC return (up vs down days)
|
||||
print(f"\n{'='*70}")
|
||||
print(f" EDGE BY BTC DAILY RETURN (SHORT)")
|
||||
print(f"{'='*70}")
|
||||
df_s['btc_up'] = df_s['btc_return'] > 0
|
||||
for up, grp in df_s.groupby('btc_up'):
|
||||
label = 'BTC UP day' if up else 'BTC DN day'
|
||||
e_vals = grp['edge'].dropna()
|
||||
print(f" {label:<15}: mean_edge={e_vals.mean():>+.2f}pp n={len(e_vals)}")
|
||||
|
||||
# Year breakdown with ExF summary
|
||||
print(f"\n{'='*70}")
|
||||
print(f" PER-YEAR: ExF MEDIANS vs EDGE")
|
||||
print(f"{'='*70}")
|
||||
print(f" {'Year':<6} {'edge_S':>8} {'edge_L':>8} {'funding':>10} {'dvol':>8} {'fng':>6} {'taker':>8} {'rvol':>10}")
|
||||
print(f" {'-'*68}")
|
||||
for yr in ['2021','2022','2023','2024','2025','2026']:
|
||||
yr_s = df_rows[(df_rows['year']==yr) & (df_rows['direction']=='S') & (df_rows['n_trades']>0)]
|
||||
yr_l = df_rows[(df_rows['year']==yr) & (df_rows['direction']=='L') & (df_rows['n_trades']>0)]
|
||||
es = yr_s['edge'].mean() if len(yr_s) > 0 else float('nan')
|
||||
el = yr_l['edge'].mean() if len(yr_l) > 0 else float('nan')
|
||||
fund = yr_s['funding_btc'].median() if yr_s['funding_btc'].notna().any() else float('nan')
|
||||
dv = yr_s['dvol_btc'].median() if yr_s['dvol_btc'].notna().any() else float('nan')
|
||||
fg = yr_s['fng'].median() if yr_s['fng'].notna().any() else float('nan')
|
||||
tk = yr_s['taker'].median() if yr_s['taker'].notna().any() else float('nan')
|
||||
rv = yr_s['realized_vol'].median()
|
||||
print(f" {yr:<6} {es:>+7.2f}pp {el:>+7.2f}pp {fund:>10.4f} {dv:>8.1f} {fg:>6.1f} {tk:>8.4f} {rv:>10.6f}")
|
||||
|
||||
# Same for LONG
|
||||
df_l = df_rows[(df_rows['direction'] == 'L') & (df_rows['n_trades'] > 0)].copy()
|
||||
print(f"\n{'='*70}")
|
||||
print(f" CORRELATIONS WITH LONG EDGE (n={len(df_l)} days with trades)")
|
||||
print(f"{'='*70}")
|
||||
for col in ['funding_btc', 'dvol_btc', 'fng', 'taker', 'btc_return', 'realized_vol', 'vd_p50']:
|
||||
sub = df_l[[col, 'edge']].dropna()
|
||||
if len(sub) < 10:
|
||||
print(f" {col:<18}: n={len(sub)} (insufficient)")
|
||||
continue
|
||||
corr = sub[col].corr(sub['edge'])
|
||||
print(f" {col:<18}: r={corr:>+.4f} (n={len(sub)})")
|
||||
|
||||
# Save
|
||||
LOG_DIR.mkdir(exist_ok=True)
|
||||
ts = datetime.now().strftime("%Y%m%d_%H%M%S")
|
||||
regime_path = LOG_DIR / f"exf_regime_{ts}.csv"
|
||||
df_rows.to_csv(regime_path, index=False)
|
||||
|
||||
# Summary JSON
|
||||
summary = {
|
||||
'timestamp': ts,
|
||||
'n_dates': total,
|
||||
'ctrl_dn_pct': round(ctrl_dn_pct, 4),
|
||||
'ctrl_up_pct': round(ctrl_up_pct, 4),
|
||||
'correlations_short': {},
|
||||
'correlations_long': {},
|
||||
'exf_availability': {},
|
||||
}
|
||||
for col in ['funding_btc', 'dvol_btc', 'fng', 'taker', 'btc_return', 'realized_vol']:
|
||||
sub_s = df_s[[col, 'edge']].dropna()
|
||||
sub_l = df_l[[col, 'edge']].dropna()
|
||||
summary['correlations_short'][col] = round(sub_s[col].corr(sub_s['edge']), 4) if len(sub_s) >= 10 else None
|
||||
summary['correlations_long'][col] = round(sub_l[col].corr(sub_l['edge']), 4) if len(sub_l) >= 10 else None
|
||||
n_avail = df_rows[df_rows['direction']=='S'][col].notna().sum()
|
||||
summary['exf_availability'][col] = int(n_avail)
|
||||
|
||||
json_path = LOG_DIR / f"exf_regime_{ts}.json"
|
||||
with open(json_path, 'w') as f:
|
||||
json.dump(summary, f, indent=2)
|
||||
|
||||
print(f"\n regime CSV → {regime_path}")
|
||||
print(f" summary → {json_path}")
|
||||
print(f" Runtime: {elapsed:.0f}s Rows: {len(df_rows):,}")
|
||||
Reference in New Issue
Block a user