- Every station's auto labels reviewed by an analyst (/review-auto-labels), then a cross-review per family across all stations (new skill /cross-review-labels), a finishing pass, and descriptions rewritten in words (no numeric values). - The three never-labeled ranges (WILDSTRUBEL 07/09-01/10, NORDEND 16/08-01/10) labeled and reviewed: 991 auto labels on 11 stations, labeling/controle.py at 0 errors. - Labeling knowledge gathered in labeling/: regles-d-etiquetage.md (new), familles.md (settled boundaries), connaissances/ (notes moved from docs/, links updated), the label map rebuilt from reviewed labels, events and cases; events tied to their stations (PS_DES_COUDRES is JUNGFRAU). The labeler and reviewers read only labeling/. - labels-map reads auto labels and events; decision D-130; living plan points 73-75. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
68 lines
4.0 KiB
Python
68 lines
4.0 KiB
Python
"""Bruit rapide des courbes corrigées, heure par heure et jour par jour, pour une station.
|
|
|
|
Bruit d'une cible à une heure : médiane des écarts de chaque mesure acceptée au milieu de ses deux voisines
|
|
(trois mesures en moins de 3 cadences ; une dérive lente, thermique, s'y annule),
|
|
corr_x et corr_y ensemble ; bruit de la station : médiane sur les cibles. Profil du lot par heure UTC ;
|
|
puis, par jour, le bruit des 6 heures les plus bruitées du profil face aux 6 plus calmes (rapport),
|
|
et si une étiquette Visibility de labeling/auto/ couvre la moitié des heures bruitées ce jour-là.
|
|
|
|
Usage (depuis la racine) : python3 bruit.py STATION DEBUT FIN [--propre] [--cibles]
|
|
"""
|
|
import sys, json, statistics as st
|
|
from datetime import timedelta
|
|
sys.path.insert(0, "agent-tools")
|
|
import donnees as d
|
|
|
|
st_nom, t1, t2 = sys.argv[1], d.instant(sys.argv[2]), d.instant(sys.argv[3])
|
|
PROPRE = "--propre" in sys.argv # retire à chaque instant la part commune à toutes les cibles (médiane signée) :
|
|
# reste le bruit propre de chaque cible (turbulence) ; une pose mal compensée, commune, s'en va
|
|
par = d.charger(f"data/{st_nom}")
|
|
cib = {s: d.filtrer(ms, t1, t2) for s, ms in par.items() if not d.est_reference(ms)}
|
|
ts = sorted({m["t"] for ms in cib.values() for m in ms})
|
|
cad = st.median([(b - a).total_seconds() for a, b in zip(ts, ts[1:])])
|
|
ecarts = {} # (cible, instant) -> (ex, ey) signés
|
|
for s, ms in cib.items():
|
|
ok = [m for m in ms if m["corr_x"] is not None]
|
|
for a, b, c in zip(ok, ok[1:], ok[2:]):
|
|
if (c["t"] - a["t"]).total_seconds() > 3 * cad:
|
|
continue
|
|
ecarts[(s, b["t"])] = tuple(b[q] - (a[q] + c[q]) / 2 for q in ("corr_x", "corr_y")) # une dérive lente s'annule
|
|
commun = {}
|
|
if PROPRE:
|
|
par_t = {}
|
|
for (s, t), e in ecarts.items():
|
|
par_t.setdefault(t, []).append(e)
|
|
commun = {t: (st.median(e[0] for e in l), st.median(e[1] for e in l)) for t, l in par_t.items() if len(l) >= 3}
|
|
cell = {} # (jour, heure) -> {cible: [|écart|]}
|
|
for (s, t), e in ecarts.items():
|
|
c0 = commun.get(t, (0, 0))
|
|
cell.setdefault((t.date(), t.hour), {}).setdefault(s, []).extend([abs(e[0] - c0[0]), abs(e[1] - c0[1])])
|
|
val = {k: st.median([st.median(v) for v in c.values()]) for k, c in cell.items() if len(c) >= 3}
|
|
prof = {h: st.median([v for (j, hh), v in val.items() if hh == h]) for h in range(24)
|
|
if any(hh == h for (j, hh) in val)}
|
|
print(f"{st_nom} cadence {cad/60:.0f} min, {len(cib)} cibles")
|
|
print("profil (heure UTC: bruit médian, unités de corr) :")
|
|
print(" " + " ".join(f"{h:02d}:{v:.3f}" for h, v in sorted(prof.items())))
|
|
ordre = sorted(prof, key=prof.get)
|
|
calme, agite = sorted(ordre[:6]), sorted(ordre[-6:])
|
|
print(f"heures calmes {calme}, agitées {agite}, rapport du profil {st.median([prof[h] for h in agite]) / st.median([prof[h] for h in calme]):.2f}")
|
|
etq = [json.loads(l) for l in open(f"labeling/auto/{st_nom}.jsonl") if l.strip()]
|
|
vis = [(d.instant(e["debut"]), d.instant(e["fin"])) for e in etq if e["categorie"] == "Visibility"]
|
|
print("jour agité calme rapport Visibility sur les heures agitées")
|
|
for j in sorted({j for j, h in val}):
|
|
a = [val[(j, h)] for h in agite if (j, h) in val]
|
|
c = [val[(j, h)] for h in calme if (j, h) in val]
|
|
if len(a) < 3 or len(c) < 3:
|
|
continue
|
|
from datetime import datetime, timezone
|
|
hs = [datetime(j.year, j.month, j.day, h, 30, tzinfo=timezone.utc) for h in agite]
|
|
couv = sum(any(a0 <= x <= b0 for a0, b0 in vis) for x in hs)
|
|
print(f"{j} {st.median(a):.3f} {st.median(c):.3f} {st.median(a) / st.median(c):5.2f} {couv}/6")
|
|
if "--cibles" in sys.argv: # rapport agité / calme de chaque cible, sur tout le lot : une gêne de la scène les touche toutes
|
|
print("cible : bruit des heures agitées / calmes (rapport)")
|
|
for s in sorted(cib):
|
|
a = [x for (j, h), c in cell.items() if h in agite for x in c.get(s, [])]
|
|
c = [x for (j, h), c in cell.items() if h in calme for x in c.get(s, [])]
|
|
if len(a) > 20 and len(c) > 20:
|
|
print(f" {s:24} {st.median(a):.4f} / {st.median(c):.4f} ({st.median(a) / max(st.median(c), 1e-6):.2f})")
|