def write():
with h5py.File(cache_path, 'a') as h5:
group = h5.require_group(result_coords)
infile_expids = set(group['expids'][:]) if 'expids' in group else {}
expidss = []
oks = 0
losses = []
foldss = []
molids = None
for expid in expids:
if verbose:
print expid, lso
if expid in infile_expids:
if verbose:
print '\tAlready done, skipping...'
continue
try:
# look for the results corresponding to the desired expid, lso
res = ManysourcesResult(expid=expid, dset=dset, feats=feats, model=model).lsocv() if lso else \
ManysourcesResult(expid=expid, dset=dset, feats=feats, model=model).crscv()
# Merge the "CV" scores to have one score per compound in the dataset
scores, labels, folds = res.merge_scores(calibration=calibration)
if verbose:
print roc_auc_score(labels, scores, average='samples')
losses.append((labels - scores) ** 2)
foldss.append(folds)
if molids is None:
molids = res.molids()
expidss.append((expid, len(infile_expids) + oks))
oks += 1
except:
# We guess that this happens when the external set only contains one class, but we need to check
print 'Warning, had troubles with', expid, lso
expidss.append((expid, -1))
# write molids - N.B. assume same for all of them, which is reasonable
if 'molids' not in group:
group['molids'] = molids
# write expids index
expids_dset = group.require_dataset('expids',
shape=(len(infile_expids) + len(expidss), 2),
dtype=np.int32,
maxshape=(None, 2))
expids_dset.resize((len(infile_expids) + len(expidss), 2))
expids_dset[len(infile_expids):] = expidss
# write losses
losses_dset = group.require_dataset('losses',
shape=(len(infile_expids) + len(losses), len(molids)),
dtype=np.float64,
maxshape=(None, len(molids)))
losses_dset.resize((len(infile_expids) + len(losses), len(molids)))
losses_dset[len(infile_expids):] = losses
# write folds (should be optional)
folds_dset = group.require_dataset('folds',
shape=(len(infile_expids) + len(losses), len(molids)),
dtype=np.int32,
maxshape=(None, len(molids)))
folds_dset.resize((len(infile_expids) + len(losses), len(molids)))
folds_dset[len(infile_expids):] = foldss
