f1b7dcf7f3
The MI sample builder used `MathAbs(labelBarOffset)` as a padding, causing rows from offset and non-offset builds to be paired with a double shift. This broke the positive control, failed the 5× gate, and voided all reported mutual‑information figures. Replace with the fixed `MiShiftPad` constant to ensure builds enumerate the same set of bars and row-k alignment is preserved. Add `BatchOptionsTotal()` to `CNeuronBatchNormOCL` and split the packed BN weight array in the learning report into separate norms for the outgoing dense matrix, gamma, beta, running statistics, and Adam moment buffers. This turns an ambiguous single‑norm reading into precise diagnostics that distinguish weight divergence from scaling issues.
92 lines
4.1 KiB
Python
92 lines
4.1 KiB
Python
"""The last claim left standing: is a broken edge a WORSE place to buy than a random level?
|
|
|
|
Everything else closed negative tonight. One sub-result survived: on EURUSD H1, a limit order
|
|
resting at the actual broken range edge did worse than the same order the same distance away
|
|
at a permuted level - `real - placebo` of -0.077 and -0.074 at phi >= 1. That is adverse
|
|
selection, and it is the mechanism book 2's A/B test proposed (a pullback deep enough to
|
|
reach the level is disproportionately a breakout that has already failed).
|
|
|
|
It is not tradeable - it is a reason NOT to do something - but it is a claim about market
|
|
structure and it was measured on one instrument. The five breadth instruments never had a
|
|
hand in finding it, so they can settle it the same way they settled the context score.
|
|
|
|
arm REAL buy limit at price_now - phi*(price_now - broken_edge)
|
|
arm PLACEBO the same distance from the same starting price, distances permuted across
|
|
events: geometry identical, level identity destroyed
|
|
|
|
Both arms pay the same spread and meet the same tie convention, so cost cancels in the
|
|
difference. The prediction, fixed in advance from the mechanism: real - placebo is NEGATIVE,
|
|
and MORE negative as phi rises, because a deeper level selects harder against you.
|
|
|
|
Note the sign convention: a negative number CONFIRMS the hypothesis here.
|
|
"""
|
|
import numpy as np, sys
|
|
sys.stdout.reconfigure(encoding='utf-8', errors='replace')
|
|
import fills, book, wyckoff, breadth
|
|
|
|
NATIVE = ('EURUSD', 'USDJPY', 'XAUUSD', 'SP500')
|
|
PHIS = (0.5, 1.0, 1.5)
|
|
|
|
|
|
def cells(tfs):
|
|
for tf in tfs:
|
|
for s in NATIVE:
|
|
bk = fills.Book(s)
|
|
yield s, tf, bk, book.frame(s, tf, bk), 'original'
|
|
for s in breadth.SPREAD_BP:
|
|
bk, f = breadth.get(s, tf)
|
|
yield breadth.NICE[s], tf, bk, f, 'new'
|
|
|
|
|
|
def gap(bk, f, sym, tf, phi, ev, seeds=(11, 23, 37)):
|
|
"""real minus placebo, averaging the placebo over several permutations."""
|
|
a = wyckoff.retest(sym, tf, phi, mrisk=2.0, kR=2.0, wait=40, H=200,
|
|
bk=bk, f=f, ev=ev)
|
|
if a is None:
|
|
return None
|
|
A = a['R'][a['indep']]
|
|
bs = []
|
|
for sd in seeds:
|
|
b = wyckoff.retest(sym, tf, phi, mrisk=2.0, kR=2.0, wait=40, H=200,
|
|
bk=bk, f=f, ev=ev, placebo=sd)
|
|
if b is not None:
|
|
bs.append(b['R'][b['indep']])
|
|
if not bs:
|
|
return None
|
|
B = np.concatenate(bs)
|
|
se = np.sqrt(A.var(ddof=1) / len(A) + B.var(ddof=1) / len(B))
|
|
return len(A), A.mean(), B.mean(), A.mean() - B.mean(), (A.mean() - B.mean()) / max(se, 1e-12)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
tfs = tuple(a for a in sys.argv[1:] if a in ('M15', 'H1', 'H4')) or ('H1',)
|
|
print("=== ADVERSE SELECTION AT THE BROKEN EDGE ===")
|
|
print(" prediction: real - placebo is NEGATIVE and grows more negative with phi.")
|
|
print(" a NEGATIVE number confirms the hypothesis.\n")
|
|
print(f" {'sym':>9}{'tf':>5}{'set':>10}" + "".join(f"{'phi='+str(p):>22}" for p in PHIS))
|
|
print(f" {'':>24}" + "".join(f"{'gap':>13}{'t':>9}" for p in PHIS))
|
|
acc = {p: {'original': [], 'new': []} for p in PHIS}
|
|
for s, tf, bk, f, kind in cells(tfs):
|
|
ev = wyckoff.breakouts(f, score=False)
|
|
if ev is None:
|
|
continue
|
|
line = f" {s:>9}{tf:>5}{kind:>10}"
|
|
any_ = False
|
|
for p in PHIS:
|
|
g = gap(bk, f, s, tf, p, ev)
|
|
if g is None:
|
|
line += f"{'-':>13}{'-':>9}"; continue
|
|
n, a, b, d, t = g
|
|
acc[p][kind].append(d)
|
|
any_ = True
|
|
line += f"{d:>+13.4f}{t:>+9.2f}"
|
|
if any_:
|
|
print(line)
|
|
print()
|
|
for p in PHIS:
|
|
o, nw = np.array(acc[p]['original']), np.array(acc[p]['new'])
|
|
al = np.concatenate([o, nw])
|
|
print(f" phi={p}: original {o.mean():+.4f} ({int((o<0).sum())}/{len(o)} negative)"
|
|
f" NEW {nw.mean():+.4f} ({int((nw<0).sum())}/{len(nw)} negative)"
|
|
f" all {al.mean():+.4f} ({int((al<0).sum())}/{len(al)})")
|
|
print("\n The NEW column is out of sample: those instruments had no hand in finding this.")
|