27 lines
1.6 KiB
Python
27 lines
1.6 KiB
Python
"""Original query metrics and paired component bootstrap, seed1 minus seed0."""
|
|
from phase2_b13_e1b import pooled
|
|
|
|
METRICS = ['relative_l2','last_location','kendall_tau_b']
|
|
|
|
|
|
def comparisons(rows):
|
|
raw=pooled([dict(component_id=r['component_id'],new=r['seed1'],A=r['seed0']) for r in rows])
|
|
out={}
|
|
for name,m in raw.items():
|
|
undefined={arm:[r['case'] for r in rows if r[arm][name]['value'] is None] for arm in ['seed0','seed1']}
|
|
out[name]=dict(means={'seed1':m['means']['new'],'seed0':m['means']['A']},
|
|
components={'seed1':m['components']['new'],'seed0':m['components']['A']},
|
|
seed1_minus_seed0=m['new_minus_A'],undefined_parts=undefined,
|
|
part_counts={arm:len(rows)-len(v) for arm,v in undefined.items()})
|
|
return out
|
|
|
|
|
|
def summary(rows):
|
|
if len(rows)!=552 or len({r['case'] for r in rows})!=552 or len({r['component_id'] for r in rows})!=204 or {r['fold'] for r in rows}!=set(range(5)):
|
|
raise ValueError('Full552/204 support required; no partial cohort inference')
|
|
return dict(parts=552,components=204,pooled=comparisons(rows),
|
|
per_fold={str(f):dict(parts=sum(r['fold']==f for r in rows),components=len({r['component_id'] for r in rows if r['fold']==f}),metrics=comparisons([r for r in rows if r['fold']==f])) for f in range(5)},
|
|
performance_gate=None,descriptive_only=True,
|
|
interval_interpretation='Fixed-model component sampling variability, not a population distribution over training seeds',
|
|
direction='seed1 minus seed0; negative field/last_location means lower error; positive tau means higher rank agreement')
|