if l_in[0].get_value() != 0. and aes_in == []:
raise ValueError("You setup the l_in but no aes in found.")
if l_out[0].get_value() != 0. and aes_out == []:
raise ValueError("You setup the l_out but no aes out found.")
# Train criterion
cost_type = CostType.MeanSquared # CostType.MeanSquared
# Compile the functions
# Momentum(0.9, nesterov_momentum=False,
# imagenet=False, imagenetDecay=5e-4,
# max_colm_norm=False)
train_updates, eval_fn = theano_fns(
model, aes_in, aes_out, l_in, l_out, l_sup, l_code, lr,
cost_type,
updaters={
"all": Momentum(0.9, nesterov_momentum=False,
imagenet=False, imagenetDecay=5e-4,
max_colm_norm=False),
"in": Momentum(0.9, nesterov_momentum=False,
imagenet=False, imagenetDecay=5e-4,
max_colm_norm=False),
"out": Momentum(0.9, nesterov_momentum=False,
imagenet=False, imagenetDecay=5e-4,
max_colm_norm=False),
"code": None},
max_colm_norm=False, max_norm=15.0, eye=False)
# How to update the weight costs
updater_wc = StaticAnnealedWeightRate(anneal_end=500, anneal_start=0)
updater_wc_in = StaticAnnealedWeightRateSingle(anneal_end=500, down=True,
init_vl=0., end_vl=0.,
anneal_start=100)
# max_colm_norm=False)
in3D = True
train_updates, eval_fn = theano_fns(model,
aes_in,
aes_out,
l_in,
l_out,
l_sup,
l_code,
lr,
cost_type,
updaters={
"all":
Momentum(0.9,
nesterov_momentum=False,
imagenet=False,
imagenetDecay=5e-4,
max_colm_norm=False),
"in":
Momentum(0.9,
nesterov_momentum=False,
imagenet=False,
imagenetDecay=5e-4,
max_colm_norm=False),
"out":
Momentum(0.9,
nesterov_momentum=False,
imagenet=False,
imagenetDecay=5e-4,
max_colm_norm=False),
"code":