167 lines
7.1 KiB
Python
167 lines
7.1 KiB
Python
|
# -*- coding: utf-8 -*-
|
||
|
|
"""P3.2.2 FORMAL EVENT DE-OVERLAP AUDIT — SniperGold_ML.
|
||
|
|
|
||
|
|
Diagnostic ONLY. No model training, no TP/SL optimization.
|
||
|
|
|
||
|
|
Formal event clustering of the EXISTING SMC event stream (no semantics change):
|
||
|
|
* overlap window H = outcome horizon (8 / 16 / 24 bars) — robustness, no "best H"
|
||
|
|
* cluster = maximal run of events with gap <= H
|
||
|
|
* lead event = earliest eligible event of each cluster (kept)
|
||
|
|
* FOLLOW_ON = all later events of a cluster (excluded from the primary
|
||
|
|
de-overlapped dataset, but never deleted — reported separately)
|
||
|
|
* verify: next_selected - selected > H for the de-overlapped stream
|
||
|
|
|
||
|
|
Per family (E_BUY / E_SELL / E_EQH / E_EQL / strict variants) and per H:
|
||
|
|
* original / clusters / lead / follow-on counts
|
||
|
|
* median & max cluster size, retention rate, follow-on rate
|
||
|
|
* ALL vs LEAD outcome comparison (WIN/LOSS/UNRESOLVED/AMBIGUOUS, MFE, MAE,
|
||
|
|
median time-to-TP, median time-to-SL) under the primary candidate combo
|
||
|
|
1.5/0.75 ATR and the symmetric sensitivity 1.0/1.0 ATR.
|
||
|
|
|
||
|
|
Entry semantics: close of the closed decision bar (unchanged from P3.2.1).
|
||
|
|
"""
|
||
|
|
import os
|
||
|
|
import sys
|
||
|
|
import datetime as dt
|
||
|
|
|
||
|
|
import numpy as np
|
||
|
|
|
||
|
|
HERE = os.path.dirname(os.path.abspath(__file__))
|
||
|
|
if HERE not in sys.path:
|
||
|
|
sys.path.insert(0, HERE)
|
||
|
|
import p3_common as P3
|
||
|
|
import label_event_audit as LEA
|
||
|
|
|
||
|
|
HORIZONS = [8, 16, 24]
|
||
|
|
COMBOS = [(1.5, 0.75), (1.0, 1.0)] # primary candidate + symmetric sensitivity
|
||
|
|
FAMILIES = ["E_BUY", "E_SELL", "E_EQH", "E_EQL",
|
||
|
|
"E_BUY_STRICT", "E_SELL_STRICT"]
|
||
|
|
|
||
|
|
|
||
|
|
def cluster_events(E, H):
|
||
|
|
"""Cluster events with gap <= H. Returns (lead, follow_on, clusters)."""
|
||
|
|
if len(E) == 0:
|
||
|
|
return np.array([], dtype=int), np.array([], dtype=int), []
|
||
|
|
leads = []
|
||
|
|
clusters = []
|
||
|
|
cur = [E[0]]
|
||
|
|
for e in E[1:]:
|
||
|
|
if e - cur[-1] <= H:
|
||
|
|
cur.append(e)
|
||
|
|
else:
|
||
|
|
clusters.append(np.array(cur, dtype=int))
|
||
|
|
leads.append(cur[0])
|
||
|
|
cur = [e]
|
||
|
|
clusters.append(np.array(cur, dtype=int))
|
||
|
|
leads.append(cur[0])
|
||
|
|
leads = np.array(leads, dtype=int)
|
||
|
|
follow = np.concatenate([c[1:] for c in clusters]) if len(clusters) else np.array([], dtype=int)
|
||
|
|
return leads, follow, clusters
|
||
|
|
|
||
|
|
|
||
|
|
def outcome_stats(E, H, combo, h, l, c, A, direction):
|
||
|
|
"""WIN/LOSS/UNRESOLVED/AMBIGUOUS + MFE/MAE + median times for event set."""
|
||
|
|
if len(E) == 0:
|
||
|
|
return None
|
||
|
|
E = E[E + H < len(c)]
|
||
|
|
if len(E) == 0:
|
||
|
|
return None
|
||
|
|
ent = c[E]
|
||
|
|
Ae = np.maximum(A[E], 1e-12)
|
||
|
|
mfe, mae, t_mfe, t_mae = LEA.mfe_mae_times(ent, E, H, Ae, h, l, direction)
|
||
|
|
tp_j, sl_j = LEA.tp_sl_scan(ent, E, H, combo[0], combo[1], Ae, h, l, direction)
|
||
|
|
oc = LEA.classify_outcome(tp_j, sl_j)
|
||
|
|
n = len(oc)
|
||
|
|
t_tp = tp_j[np.isfinite(tp_j)]
|
||
|
|
t_sl = sl_j[np.isfinite(sl_j)]
|
||
|
|
return {
|
||
|
|
"n": int(n),
|
||
|
|
"P_win": float((oc == "WIN").mean()),
|
||
|
|
"P_loss": float((oc == "LOSS").mean()),
|
||
|
|
"P_unresolved": float((oc == "UNRESOLVED").mean()),
|
||
|
|
"P_ambiguous": float((oc == "AMBIGUOUS").mean()),
|
||
|
|
"MFE_median_atr": float(np.nanmedian(mfe)),
|
||
|
|
"MAE_median_atr": float(np.nanmedian(mae)),
|
||
|
|
"median_t_TP": float(np.median(t_tp)) if len(t_tp) else None,
|
||
|
|
"median_t_SL": float(np.median(t_sl)) if len(t_sl) else None,
|
||
|
|
}
|
||
|
|
|
||
|
|
|
||
|
|
def main():
|
||
|
|
t, o, h, l, c, v, htf = P3.load_data()
|
||
|
|
F = P3.load_F()
|
||
|
|
A = P3.atr_series(h, l, c)
|
||
|
|
n = len(c)
|
||
|
|
lab = P3.make_label(c, A)
|
||
|
|
idx, direction = LEA.event_masks(F, n)
|
||
|
|
F8 = F[:, 8]
|
||
|
|
F5 = F[:, 5]
|
||
|
|
|
||
|
|
provenance = P3.provenance()
|
||
|
|
provenance["P3_1_P3_2_1_CHECKPOINT_SHA"] = "dc1faa92f36ad22d4062315a5654965a08a006dc"
|
||
|
|
provenance["script_sha"] = P3.sha256_file(os.path.abspath(__file__))
|
||
|
|
provenance["dataset_hash"] = P3.dataset_hash(F, lab, A, t)
|
||
|
|
provenance["event_semantics"] = "existing FEATURE_CONTRACT v1.0 (f9/f7/f10/f11)"
|
||
|
|
provenance["entry_semantics"] = "close of closed decision bar"
|
||
|
|
provenance["invalidation_semantics"] = "opposite structure break (f8 or f5 flip)"
|
||
|
|
|
||
|
|
print("=== P3.2.2 FORMAL DE-OVERLAP AUDIT ===")
|
||
|
|
for H in HORIZONS:
|
||
|
|
report = {"provenance": provenance,
|
||
|
|
"overlap_window_H": H, "combo": [list(x) for x in COMBOS],
|
||
|
|
"families": {}}
|
||
|
|
print("\n--- H=%d ---" % H)
|
||
|
|
for k in FAMILIES:
|
||
|
|
E = idx[k]
|
||
|
|
d = direction[k]
|
||
|
|
leads, follow, clusters = cluster_events(E, H)
|
||
|
|
# verify de-overlap
|
||
|
|
ok_verify = bool(np.all(np.diff(leads) > H)) if len(leads) > 1 else True
|
||
|
|
sizes = np.array([len(c_) for c_ in clusters]) if clusters else np.array([0])
|
||
|
|
fam = {
|
||
|
|
"n_original": int(len(E)),
|
||
|
|
"n_clusters": int(len(clusters)),
|
||
|
|
"n_lead": int(len(leads)),
|
||
|
|
"n_follow_on": int(len(follow)),
|
||
|
|
"median_cluster_size": float(np.median(sizes)) if len(sizes) else None,
|
||
|
|
"max_cluster_size": float(np.max(sizes)) if len(sizes) else None,
|
||
|
|
"retention_rate": float(len(leads) / len(E)) if len(E) else None,
|
||
|
|
"follow_on_rate": float(len(follow) / len(E)) if len(E) else None,
|
||
|
|
"deoverlap_verified_next_gt_H": bool(ok_verify),
|
||
|
|
"outcomes": {},
|
||
|
|
}
|
||
|
|
for combo in COMBOS:
|
||
|
|
key = "%s/%s" % (combo[0], combo[1])
|
||
|
|
fam["outcomes"][key] = {
|
||
|
|
"ALL": outcome_stats(E, H, combo, h, l, c, A, d),
|
||
|
|
"LEAD": outcome_stats(leads, H, combo, h, l, c, A, d),
|
||
|
|
}
|
||
|
|
report["families"][k] = fam
|
||
|
|
oa = fam["outcomes"]["1.5/0.75"]
|
||
|
|
print(" %-14s orig=%6d lead=%5d follow=%6d ret=%.4f med_cl=%5.0f "
|
||
|
|
"max_cl=%5.0f verify=%s" % (
|
||
|
|
k, fam["n_original"], fam["n_lead"], fam["n_follow_on"],
|
||
|
|
fam["retention_rate"], fam["median_cluster_size"],
|
||
|
|
fam["max_cluster_size"], ok_verify))
|
||
|
|
if oa["ALL"] and oa["LEAD"]:
|
||
|
|
print(" ALL : W=%.3f L=%.3f U=%.3f | MFE=%.2f MAE=%.2f | "
|
||
|
|
"tTP=%s tSL=%s" % (
|
||
|
|
oa["ALL"]["P_win"], oa["ALL"]["P_loss"],
|
||
|
|
oa["ALL"]["P_unresolved"], oa["ALL"]["MFE_median_atr"],
|
||
|
|
oa["ALL"]["MAE_median_atr"],
|
||
|
|
("%.1f" % oa["ALL"]["median_t_TP"]) if oa["ALL"]["median_t_TP"] else "-",
|
||
|
|
("%.1f" % oa["ALL"]["median_t_SL"]) if oa["ALL"]["median_t_SL"] else "-"))
|
||
|
|
print(" LEAD: W=%.3f L=%.3f U=%.3f | MFE=%.2f MAE=%.2f | "
|
||
|
|
"tTP=%s tSL=%s" % (
|
||
|
|
oa["LEAD"]["P_win"], oa["LEAD"]["P_loss"],
|
||
|
|
oa["LEAD"]["P_unresolved"], oa["LEAD"]["MFE_median_atr"],
|
||
|
|
oa["LEAD"]["MAE_median_atr"],
|
||
|
|
("%.1f" % oa["LEAD"]["median_t_TP"]) if oa["LEAD"]["median_t_TP"] else "-",
|
||
|
|
("%.1f" % oa["LEAD"]["median_t_SL"]) if oa["LEAD"]["median_t_SL"] else "-"))
|
||
|
|
P3.save_json("p3_2_deoverlap_h%d.json" % H, report)
|
||
|
|
|
||
|
|
print("\nP3.2.2 de-overlap audit selesai: p3_2_deoverlap_h8/16/24.json")
|
||
|
|
|
||
|
|
|
||
|
|
if __name__ == "__main__":
|
||
|
|
main()
|