Datasets:
Tasks:
Other
Formats:
parquet
Languages:
English
Size:
1K - 10K
Tags:
machine-learning
active-learning
representation-learning
sparse-fourier-transform
experimental-design
synthetic-data
File size: 4,822 Bytes
5b2d580 | 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 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 | """Validate the prepared dataset locally; never authenticates or uploads."""
from pathlib import Path
import json,os,hashlib
os.environ['HF_HUB_OFFLINE']='1'
os.environ['HF_DATASETS_OFFLINE']='1'
import pyarrow.parquet as pq
import yaml,datasets,huggingface_hub
from huggingface_hub import DatasetCard
ROOT=Path(__file__).resolve().parents[1]
def main():
front=ROOT.joinpath('README.md').read_text(encoding='utf-8').split('---',2)[1]
meta=yaml.safe_load(front)
assert len(meta['configs'])==8
assert sum(bool(c.get('default'))for c in meta['configs'])==1
assert meta['configs'][0]['config_name']=='evaluation'
assert 'license'not in meta
card=DatasetCard.load(str(ROOT/'README.md'))
assert len(card.data.to_dict()['configs'])==8
source={};records=[];contexts=[];row_counts={}
for p in sorted((ROOT/'data').glob('*.parquet')):
rows=pq.read_table(p).to_pylist()
mirrors=[json.loads(l)for l in (ROOT/'tables_jsonl'/f'{p.stem}.jsonl').read_text().splitlines()]
assert rows==mirrors
row_counts[p.stem]=len(rows)
if p.stem=='contexts':contexts=rows
else:records.extend(rows)
context_map={r['context_id']:r for r in contexts}
assert len(context_map)==74
total_masks=0;large_masks=0
for r in records:
f=r['source_file']
if f not in source:source[f]=json.loads((ROOT/f).read_text())
obj=source[f]
for token in r['source_pointer'].split('/')[1:]:
token=token.replace('~1','/').replace('~0','~')
obj=obj[int(token)]if isinstance(obj,list)else obj[token]
truth=obj['truth_evaluation']
assert r['oracle_calls']==obj['total_oracle_calls']
assert r['component_mse']==truth['component_mse']
assert r['certified']==truth['certified']
assert r['false_certificate_evaluation_only']==truth['false_certificate_evaluation_only']
assert r['oracle_calls']==r['specialized_readout_calls']+r['fresh_risk_audit_calls']+r['search_and_screen_calls']
assert r['context_id']in context_map
for c in json.loads(r['charged_to_context_ids_json']):assert c in context_map
masks=r['model_support_masks'];coefs=r['model_coefficients']
assert len(masks)==len(coefs)==r['model_terms']
assert dict(zip(masks,coefs))==obj['model']
for mask in masks:
assert isinstance(mask,str)and str(int(mask))==mask
assert 0<=int(mask)<(1<<r['dimension'])
large_masks+=int(int(mask)>2**64-1);total_masks+=1
evaluations=[r for r in records if r['phase']=='evaluation']
warmup=[r for r in records if r['phase']=='warmup']
assert len(evaluations)==668 and len(warmup)==41
assert len({r['record_id']for r in records})==709
assert len({r['teacher_case_id']for r in evaluations})==148
assert len({r['teacher_case_id']for r in warmup})==21
assert sum(r['certified']for r in evaluations)==661
for c in contexts:
calls=sum(r['oracle_calls']for r in evaluations if r['context_id']==c['context_id'])
assert c['evaluation_oracle_calls']==calls
assert c['training_inclusive_calls']==calls+c['charged_training_calls']
expected={'evaluation':668,'core_evaluation':324,'probe_evaluation':120,'strong_evaluation':192,
'unified_evaluation':32,'warmup_usages':41,'all_risk_records':709,'contexts':74}
loaded={}
# Load the local repository using actual dataset-card configuration, not
# an independently reconstructed list of files.
for config in meta['configs']:
name=config['config_name']
data=datasets.load_dataset(str(ROOT),name=name)
assert len(data)==1
split=list(data)[0]
assert data[split].num_rows==expected[name]
loaded[name]={'split':split,'rows':data[split].num_rows,'columns':len(data[split].column_names)}
default=datasets.load_dataset(str(ROOT))
assert default['evaluation'].num_rows==668
archive=ROOT/'release/Nhioryum_Phi_Research_Package.zip'
sha=hashlib.sha256(archive.read_bytes()).hexdigest()
assert sha=='925b1b7a917c5a16c25f1ade5a08c9622366cbe3b0e6da828a3e9fbfa7ca8e54'
result={'local_card_metadata_valid':True,'actual_datasets_local_load':loaded,
'default_config_load_rows':668,'parquet_jsonl_exact_roundtrip':True,
'source_pointer_records_verified':len(records),'context_rows_verified':len(contexts),
'support_masks_verified':total_masks,'support_masks_above_uint64_preserved_as_strings':large_masks,
'original_archive_sha256':sha,'datasets_version':datasets.__version__,
'huggingface_hub_version':huggingface_hub.__version__,
'remote_upload_performed':False,'network_disabled_for_validation':True}
print(json.dumps(result,indent=2))
if __name__=='__main__':main()
|