54 lines
3.7 KiB
Python
54 lines
3.7 KiB
Python
"""Metadata-only Track A postfit binding while the original training continues."""
|
|
from pathlib import Path
|
|
import time
|
|
from phase2_h1_worker_v2 import sha, checked, save
|
|
from track_a_h2_common import config as training_config, read
|
|
from track_a_h2_prepare import local_code_closure
|
|
|
|
B=Path(__file__).resolve().parents[1]
|
|
T=Path('/home/frank14f/MeltMind-runs/track-a-h2-2026-11-18')
|
|
R=T.parent/'track-a-postfit-2026-11-18'
|
|
|
|
|
|
def main():
|
|
start=time.process_time()
|
|
if R.exists():raise FileExistsError('Existing postfit entry; inspect rather than duplicate')
|
|
tc,_=training_config(T);entry=read(tc['reuse_entry']);sources=dict(tc['source_sha256'])
|
|
for p in [T/'contract.json',T/'implementation-git-freeze.json',T/'implementation-authorization.json',T/'review.json']:
|
|
sources[str(p)]=sha(p)
|
|
frozen=read(checked(entry['A_predictions_freeze']['path'],entry['A_predictions_freeze']['sha256']))
|
|
originals={(r['split'],r['case']):r for r in frozen['rows']};A={};count=0
|
|
for j in tc['jobs']:
|
|
index=j['id'];A[str(index)]={}
|
|
for case in j['test']:
|
|
row=originals[(index,case)];ref=row['predictions']['GBRT']
|
|
if row['status']!='PREDICTED' or row['raw_ids_sha256']!=tc['parts'][case]['output_sha256']['raw_ids.npy'] or ref['nodes']!=tc['parts'][case]['nodes'] or not ref['no_clipping']:raise ValueError('Original A prediction/native-ID identity')
|
|
A[str(index)][case]=dict(ref,raw_ids_sha256=row['raw_ids_sha256'])
|
|
count+=1
|
|
expected_unique=len(set().union(*(set(j['test']) for j in tc['jobs'])))
|
|
if count!=1537 or expected_unique!=586:raise ValueError('Original paired support differs')
|
|
codes=local_code_closure([B/'tools'/n for n in ['track_a_postfit.py','track_a_postfit_common.py',
|
|
'track_a_postfit_worker.py','track_a_postfit_guard.py','track_a_postfit_statistics.py',
|
|
'prepare_track_a_postfit.py','test_track_a_postfit.py']])
|
|
for p,h in tc['code_sha256'].items():checked(p,h)
|
|
c=dict(version='track-a-postfit-v1',approval='1794975970.156219',budget_approval='1794982020.974879',estimator_approval='1794982202.809969',
|
|
training_root=str(T),training_contract_sha256=sha(T/'contract.json'),
|
|
comparison_contract=tc['reuse_entry'],comparison_contract_sha256=sha(tc['reuse_entry']),
|
|
source_sha256=sources,code_sha256=codes,jobs=[dict(id=j['id'],expected_parts=len(j['test'])) for j in tc['jobs']],
|
|
A_predictions=A,expected_pair_visits=count,expected_unique_target_reads=expected_unique,
|
|
gpu_uuid='GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76',worker_cpus=[4,5],coordinator_cpus=[12,13],
|
|
gpu_cap=162000,postfit_gpu_cap=7200,postfit_cpu_cap=3600,idle_wait_seconds=20,no_retry=True,
|
|
bootstrap=dict(replicates=2000,seed=2026111408,confidence=.95,unit='unique_part_mean_of_split_visit_differences',interval='percentile_linear'),
|
|
no_auto_launch=True,all_five_training_terminals_before_prediction=True,all_predictions_frozen_before_targets=True,
|
|
A_refit=False,H2_refit=False,new_decodes=0,metric=tc['metric'],original_failed=tc['original_failed'])
|
|
R.mkdir();save(R/'contract.json',c)
|
|
save(R/'preparation-receipt.json',dict(status='PREPARED_PENDING_EXACT_PRE_SCORE_REVIEW',
|
|
contract_sha256=sha(R/'contract.json'),code_files=len(codes),paired_visits=count,unique_cases=expected_unique,
|
|
metadata_cpu_seconds=time.process_time()-start,source_training_unchanged=True,model_weights_read=False,
|
|
scientific_arrays_read=False,target_values_read=False,old_scores_read=False,new_predictions=False,
|
|
training_complete_exists=(T/'complete.json').exists(),launched=False))
|
|
print(read(R/'preparation-receipt.json'))
|
|
|
|
|
|
if __name__=='__main__':main()
|