auto: the analysts verified the labels of 32 periods (11 stations, up to 14 days each): 710 labels -> 835, 587 unchanged; one analysis folder per period with its report (section Verification)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
1 parent
df979301b5
commit
2a5d87659b
279 files changed
+10758
-710
No files matched your search
@@ -0,0 +1,28 @@
|
||||
"""Builds labels.jsonl from work/existing.jsonl: kept labels unchanged (review kept, id dropped), corrections, additions."""
|
||||
import json
|
||||
D = 'labeling/auto/analyses/NORDEND_2026-09-04T0615_2026-09-17T0545/'
|
||||
old = {l['id']: l for l in map(json.loads, open(D + 'work/existing.jsonl'))}
|
||||
fix = {
|
||||
'NORDEND-101': {'end': '2026-09-08T02:45Z', 'description': "At night, PR07 is rejected for two measurements: its curve is missing; at the same time PR16 jumps on corr_x and corr_y, and PR15 a little on corr_x, then they come back at the measurement that follows the rejections. The 02:00 image is very noisy and the weak spot of PR07 is lost in it. Points to ignore, nothing to do."},
|
||||
'NORDEND-102': {'start': '2026-09-08T05:30Z', 'end': '2026-09-08T08:15Z', 'description': "Early in the morning, the corr_x of PR07 alone drops below its level, a little at first, then clearly for more than an hour, without rejection, and comes back gradually; its corr_y rises a little and its error_value rises at the same time, without reaching the threshold. The other targets do not follow it. Its spot is weak on the 07:00 thumbnail, without visible obstacle. Cause not established: a movement of PR07 itself is not excluded, but it comes back by itself; to monitor."},
|
||||
'NORDEND-106': {'end': '2026-09-09T16:00Z', 'description': "CTR01 (local reference, treated as a target) is rejected for two measurements in a row in the afternoon: its corr_x and corr_y curves are missing. Deduced cause: same hour and same signature as the other sunny afternoons, where the thumbnails show the rock in full sun around the target and the spot drowning in it (16/08, 21 and 22/09). At the second of them, the error_value of almost all the sensors rises and several targets move a little: PR18, PR15 and PR08 on corr_y, PR14, PR05 and PR01 on corr_x, most for one measurement, PR08 and PR15 coming back on corr_y only half an hour later. The other targets remain measured. The structure does not move: nothing to do, CTR01's curve is missing on sunny afternoons."},
|
||||
'NORDEND-111': {'end': '2026-09-13T05:45Z'},
|
||||
'NORDEND-113': {'start': '2026-09-13T05:45Z'},
|
||||
'NORDEND-114': {'end': '2026-09-13T16:00Z', 'description': "In early afternoon, the corrections of several targets fluctuate, one-measurement jumps in one direction then the other, mostly PR15, PR14, PR16, PR18, and a little PR08, PR06, PR11, PR13; PR07 is rejected most of the time and PR16 once, the error_value of several targets rises. After its last rejection, PR07 comes back clearly below its level on corr_x for two measurements, its error_value still high, before returning. The camera is calm. Images from 12:00 to 14:45: nothing in front of the targets, spots a little larger towards the end. Cause not established, perhaps daytime heat shimmer. The structure does not move: do not trust these points."},
|
||||
'NORDEND-122': {'end': '2026-09-16T08:45Z', 'description': "PR07 is rejected for two measurements, its curve is missing; at the same time PR13 and PR16 rise a little on corr_y for three measurements, their raw positions departing from the overall movement of the targets, then come back. Meanwhile the housing cools for a moment and the camera turns back a little (thermal cycle, labelled separately). Nothing visible in front of the targets on the 08:00 image, the spot of PR07 is weak there. Cause not established (the cobweb seen at night in the field on those days is not seen at this hour). Points to ignore, nothing to do."},
|
||||
}
|
||||
add = [
|
||||
{'start': '2026-09-12T22:45Z', 'end': '2026-09-13T00:00Z', 'family': 'Visibility', 'description': "At night, PR07 makes a clear corr_x jump for two measurements without being rejected, after a small dip just before, and its corr_y rises a little, then it comes back; its error_value rises at the first of them, and its neighbour PR04 is rejected at the same two measurements. On the 23:15 and 23:30 thumbnails, the spots of PR07 and PR04 are smaller and fainter than at 23:00 and 23:45, nothing visible in front of them, while PR06 stays unchanged. Cause not established: a poorer view of these targets at night, as on other nights of this station (a lit cobweb thread is possible, not seen). Points to ignore, this is not a movement."},
|
||||
]
|
||||
out = []
|
||||
for i, l in old.items():
|
||||
l = {k: v for k, v in l.items() if k != 'id'}
|
||||
if i in fix:
|
||||
l.update(fix[i]); l.pop('review', None)
|
||||
out.append(l)
|
||||
out += add
|
||||
out.sort(key=lambda l: (l['start'], l['end']))
|
||||
with open(D + 'labels.jsonl', 'w') as f:
|
||||
for l in out:
|
||||
f.write(json.dumps(l, ensure_ascii=False) + '\n')
|
||||
print(len(out), 'labels')
|
||||
Reference in new issue
Block a user