def sources_coocurrences_df(dset='bcrp',
expids=None,
feats='ecfps1',
model='logreg3',
lso=True):
cache_file = op.join(MANYSOURCES_DATA_ROOT, 'results', 'sources_coocurrences_df.h5')
if expids is None:
expids = range(4096)
group_id = '/dset=%s/feats=%s/model=%s/lso=%r' % (dset, feats, model, lso)
sources_dataset_id = '/dset=%s/sources' % dset
with h5py.File(cache_file, 'a') as h5:
if group_id not in h5:
coocurrences = []
valid_expids = []
fold_ids = []
dset_name = dset
dset = None
for expid in expids:
print dset_name, expid, model, feats, lso
res = ManysourcesResult(expid=expid, dset=dset_name, feats=feats, model=model)
if dset is None:
dset = res.ms_dset()
if sources_dataset_id not in h5:
sources_as_in_the_matrix = dset.mols().i2sources_order()
h5[sources_dataset_id] = sources_as_in_the_matrix
if lso is False:
print '\tWARNING: source coocurrences do not make much sense without LSO splitting'
cv = res.lsocv() if lso else res.crscv()
if cv is None:
continue
for fold_num, fold in enumerate(cv.folds()):
try:
coocurrences.append(fold.sources_coocurrences(dset=dset))
valid_expids.append(expid)
fold_ids.append(fold_num)
except:
pass
res._close_h5()
group = h5.require_group(group_id)
group['coocurrences'] = np.array(coocurrences)
group['expids'] = valid_expids
group['folds'] = fold_ids
with h5py.File(cache_file, 'r') as h5:
sources = h5[sources_dataset_id][:]
coocurrences = h5[group_id]['coocurrences'][:].astype(np.bool)
expids = h5[group_id]['expids'][:]
folds = h5[group_id]['folds'][:]
return coocurrences, sources, expids, folds
def molecules_coocurrences_df(dset='bcrp',
expids=None,
feats='ecfps1',
model='logreg3',
lso=True):
cache_file = op.join(MANYSOURCES_DATA_ROOT, 'results', 'molecules_coocurrences_df.h5')
if expids is None:
expids = range(4096)
group_id = '/dset=%s/feats=%s/model=%s/lso=%r' % (dset, feats, model, lso)
molecules_dataset_id = '/dset=%s/molecules' % dset
with h5py.File(cache_file, 'a') as h5:
if group_id not in h5:
coocurrences = []
valid_expids = []
fold_ids = []
dset_name = dset
dset = None
for expid in expids:
print dset_name, expid, model, feats, lso
res = ManysourcesResult(expid=expid, dset=dset_name, feats=feats, model=model)
if dset is None:
dset = res.ms_dset()
if molecules_dataset_id not in h5:
molecules_as_in_the_matrix = res.molids()
h5[molecules_dataset_id] = molecules_as_in_the_matrix
cv = res.lsocv() if lso else res.crscv()
if cv is None:
continue
for fold_num, fold in enumerate(cv.folds()):
try:
c = np.zeros(len(res.molids()), dtype=np.int)
c[fold.test_indices()] = 1
coocurrences.append(c)
valid_expids.append(expid)
fold_ids.append(fold_num)
except:
pass
res._close_h5()
group = h5.require_group(group_id)
group['coocurrences'] = np.array(coocurrences)
group['expids'] = valid_expids
group['folds'] = fold_ids
with h5py.File(cache_file, 'r') as h5:
molids = h5[molecules_dataset_id][:]
coocurrences = h5[group_id]['coocurrences'][:].astype(np.bool)
expids = h5[group_id]['expids'][:]
folds = h5[group_id]['folds'][:]
return coocurrences, molids, expids, folds
