File size: 3,058 Bytes
1118fb8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 | """Verify downloaded bytes, split counts, label mapping and exact cache layout."""
import argparse,hashlib,json
from collections import Counter
from pathlib import Path
import pyarrow.parquet as pq
from release_reader import load_processed,read_frames,resolve
def sha(path):
h=hashlib.sha256()
with path.open('rb') as f:
for b in iter(lambda:f.read(1024*1024),b''):h.update(b)
return h.hexdigest()
def main(root):
files=0
listed=set()
for line in (root/'checksums.sha256').read_text().splitlines():
digest,relative=line.split(' ',1)
assert relative not in listed,relative
listed.add(relative)
assert sha(resolve(root,relative))==digest,relative
files+=1
actual={str(p.relative_to(root)) for p in root.rglob('*') if p.is_file()}
# HF download metadata and Python bytecode are local runtime artifacts.
actual={p for p in actual if not p.startswith('.cache/') and '__pycache__' not in Path(p).parts}
assert actual==listed|{'checksums.sha256'},'Unexpected or missing package files'
expected={'train':4800,'test_ind':1200,'test_ood':400}
video_splits={}
for split,count in expected.items():
rows=load_processed(root,split);assert len(rows)==count
byid={r['id']:r for r in rows};assert len(byid)==count
for row in rows:
frames=read_frames(root,row)
assert frames.shape==(32,224,224,3) and str(frames.dtype)=='uint8'
assert len(row['target_progress'])==32
total=0;seen=set();classes=Counter()
for p in sorted((root/'data'/split).glob('*.parquet')):
for batch in pq.ParquetFile(p).iter_batches(batch_size=8):
for row in batch.to_pylist():
assert row['split']==split
assert row['id'] not in seen
seen.add(row['id'])
digest=hashlib.sha256(row['video']['bytes']).hexdigest()
assert digest==row['video_sha256']==sha(resolve(root,row['repro_video']))
assert video_splits.setdefault(digest,split)==split,'Cross-split duplicate video'
assert row['processed_target_progress']==byid[row['id']]['target_progress']
assert row['lang_vector']==byid[row['id']]['lang_vector']
assert len(row['target_progress'])==len(row['progress_frame_indices'])
assert row['progress_frame_indices']==list(range(row['video_frame_offset'],row['video_num_frames']))
classes[(row['task_id'],row['quality_label'])]+=1
total+=1
assert total==count
assert seen==set(byid)
assert len(classes)==(8 if split=='test_ood' else 24)
assert set(classes.values())=={200 if split=='train' else 50}
print(json.dumps({'status':'passed','files':files,'items':sum(expected.values())}))
if __name__=='__main__':
p=argparse.ArgumentParser();p.add_argument('--root',type=Path,required=True);a=p.parse_args();main(a.root.resolve())
|