def load_factors():
factors = []
for a in range(12, 15):
for e in [1]:
animal = param.animal_list[a]
meta_df, roi_tensor, acti, norm_acti, smoothed_acti = data.load_processed_data_all(
animal)
days = set(meta_df.loc[meta_df['Experiment Class'] == e,
'Day'].tolist())
for d in days:
path = os.path.join(
paths.path2Output, animal, 'non_negative_parafac',
'random', '6',
'Day-[{}]_Experiment Class-[{}]'.format(d, e),
'factorsDay-[{}]_Experiment Class-[{}]_00.npy'.format(
d, e))
print(path)
factor = np.load(path)
#Compute norms along columns for each factor matrix
norms = [sci.linalg.norm(f, axis=0) for f in factor]
# Multiply norms across all modes
lam = sci.multiply.reduce(norms)**(1 / 3)
# Update factors
factor = [f * (lam / fn) for f, fn in zip(factor, norms)]
factors.append(factor)
return factors
def load_factors(method):
"""Load factors in order to generate global time factor
Returns
-------
list
List of factors
"""
factors = []
for a in fixed_selection['Animal']:
animal = param.animal_list[a]
meta_df, roi_tensor, acti, norm_acti, smoothed_acti = data.load_processed_data_all(
animal)
for e in fixed_selection['Experiment Class']:
days = set(meta_df.loc[meta_df['Experiment Class'] == e,
'Day'].tolist())
print(days)
for day in days:
current_fixed_selection = {
'Experiment Class': [e],
'Day': [day]
}
print(current_fixed_selection)
name = ''.join([
'_' + k + '-' + str(current_fixed_selection[k])
for k in current_fixed_selection
if not current_fixed_selection[k] == None
])[1:]
path = os.path.join(paths.path2Output, animal, method,
args.init, str(args.rank), name)
if os.path.exists(path):
factor = np.load(
os.path.join(path, 'factors' + name + '_00.npy'))
else:
os.makedirs(path)
_, acti_select, norm_acti_select, smoothed_acti_select = data.select_data(
meta_df, acti, norm_acti, smoothed_acti,
current_fixed_selection)
model = t.TCA(function=method,
rank=args.rank,
init=args.init,
verbose=args.verbose,
roi_tensor=roi_tensor)
factor = model.fit(torch.tensor(norm_acti_select))
np.save(
os.path.join(path, 'factors{}_00.npy'.format(name)),
factor)
#Compute norms along columns for each factor matrix
norms = [sci.linalg.norm(f, axis=0) for f in factor]
# Multiply norms across all modes
lam = sci.multiply.reduce(norms)**(1 / 3)
# Update factors
factor = [f * (lam / fn) for f, fn in zip(factor, norms)]
factors.append(factor)
return factors
name = ''.join([
'_' + k + '-' + str(selection[k]) for k in selection
if not selection[k] == None
])[1:]
path = os.path.join(paths.path2Output, animal, args.function, args.init,
str(args.rank), name)
path_fig = os.path.join(paths.path2Figures, animal, args.function, args.init,
str(args.rank), name)
for p in [path, path_fig]:
try:
os.makedirs(p)
except:
FileExistsError
meta_df, roi_tensor, acti, norm_acti, smoothed_acti = data.load_processed_data_all(
animal)
meta_df, acti, norm_acti, smoothed_acti = data.select_data(
meta_df, acti, norm_acti, smoothed_acti, selection)
if args.similarity:
sim = an.kruskal_align(fac1, fac2)
if args.verbose: print(sim)
if args.TSNE:
an.TSNE_trials(acti, meta_df, name, path_fig, dataset=args.TSNE)
if args.correlation:
for b in ['CR', 'HIT', 'MISS', 'FA']:
an.correlation_clustering(acti, meta_df, name, path_fig, be=b)
###¨Put predict