def data(self):
if not self.grp.get('data'): return None # no data exists
attrs = self.grp['data'].attrs
if attrs.get('reference'):
d = load_nii_or_npy(attrs['reference'])
return AttrArray(d, attrs) # data from external npy or nifti
else:
return self.grp['data'] # data from hdf5
def create_cond(self, condname, run=None, group=None, offset=0, max_len=False, threshold=0, audio_env=None,
base_dir="", nii_files=None, dry_run=False, reference=False, **kwargs):
"""Create condition.
Generally used by passing cond params from setup, but can be called manually.
All parameters set as default for each run in the condition.
Parameters:
reference: should the data be copied in, or use a reference to data instead?
"""
cond = self.f['conds'].create_group(condname)
cond.attrs['offset'] = offset
cond.attrs['max_len'] = max_len or False #can't store None
cond.attrs['threshold'] = threshold
cond.attrs['prop_pass_thresh'] = .7
cond.attrs['run'] = run or condname
cond.attrs['reference'] = reference
if group: cond.attrs['group'] = group
cond.create_group('blocks')
cond.create_group('correlations')
cond.create_group('analyses')
#Load conditions in from config
if audio_env:
aud_path = os.path.join(base_dir, audio_env)
print aud_path
cond.create_dataset('audio_env', data=load_nii_or_npy(aud_path))
#Don't attempt to load data when nii_files is blank (dummy cond)
if not nii_files: return
#Create subject data
full_query = os.path.join(base_dir, nii_files)
for m in self.get_subject_files(full_query):
fname = m.group()
sub_id = m.groupdict()['sub_id'] #TODO add run_id to mix
print fname, sub_id
if not dry_run: self.create_subrun(sub_id, condname, fname, **cond.attrs)
#have option to do threshold?
self.f.flush()
#TODO blocks (pandas)
if dry_run:
for k, v in cond.attrs.iteritems(): print k, ':\t\t', v
del self.f[cond.name]
return cond
def create_dataset(self, data, overwrite=False, reference=False, compression='gzip', chunks=(1,), **kwargs):
"""Data can be np.ndarray, nifti, or file name. Remaining dimensions for chunks are inferred from data
Even if ref is True, still loads data (to ensure it exists)
"""
#TODO should fill attributes be mandatory?
#TODO ref and overwrite args
if self.data and not overwrite:
raise BaseException('data already exists')
if type(data) is str: np_arr = load_nii_or_npy(data) #string
elif type(data) is nib.nifti1.Nifti1Image: np_arr = data.get_data() #nifti
else: np_arr = data #np array
#rdy_arr = np_arr.reshape([-1, np_arr.shape[-1]])
chunks += np_arr.shape[len(chunks):] #fill in rest of chunk information
if reference:
self.grp.create_group('data') #make dummy group for attrs
reference = data # file path
else:
self.grp.create_dataset('data', data=np_arr, chunks=chunks, compression=compression)
if kwargs: self.fill_attributes(reference=reference, **kwargs)
self.grp.file.flush()
def create_roi(self, roiname, fname):
"""Add new roi to hdf5 in /rois group"""
roi_data = load_nii_or_npy(fname)
roi = self.f['rois'].create_dataset(roiname, data=roi_data.astype(bool))
return roi