pre_file = 'incept_temporal{}_{}'.format(temp_rate, n_neurons)
if train & (not retrain):
weights = 'imagenet'
else:
weights = None
if args.fine == 1:
fine = True
else:
fine = False
result_model = models.InceptionTemporal(n_neurons=n_neurons,
seq_len=seq_len,
classes=classes,
weights=weights,
dropout=dropout,
fine=fine,
retrain=retrain,
pre_file=pre_file,
old_epochs=old_epochs,
cross_index=cross_index)
if (args.summary == 1):
result_model.summary()
sys.exit()
lr = args.lr
decay = args.decay
result_model.compile(loss='categorical_crossentropy',
optimizer=optimizers.SGD(lr=lr,
decay=decay,
model_s = models.InceptionSpatial(n_neurons=n_neurons,
seq_len=seq_len,
classes=classes,
weights=None,
dropout=dropout,
fine=False,
retrain=False,
pre_file=pre_file,
old_epochs=old_epochs,
cross_index=cross_index)
model_t = models.InceptionTemporal(n_neurons=n_neurons,
seq_len=seq_len,
classes=classes,
weights=None,
dropout=dropout,
fine=False,
retrain=False,
pre_file=pre_file,
old_epochs=old_epochs,
cross_index=cross_index)
print(glob.glob('weights/' + pre_file + '-{:2d}*.hdf5'.format(epochs))[-1])
model_s.load_weights(
glob.glob('weights/' + pre_file + '-{:2d}*.hdf5'.format(epochs))[-1])
#model_s.summary()
weights = model_s.layers[1].get_weights()
#print(len(weights))
weights_0 = weights[0]
print(np.asarray(weights_0).shape)