'attention_type': args.attention_type,
'attention_alignment': args.attention_alignment,
'encoder_type': args.encoder_type,
'weights_init': w_init,
'biases_init': b_init,
'raw_output': args.raw_output,
'name': 'parrot'}
parrot = Parrot(**parrot_args)
parrot.initialize()
features, features_mask, labels, labels_mask, speaker, start_flag, raw_sequence = \
parrot.symbolic_input_variables()
cost, extra_updates, attention_vars, cost_raw = parrot.compute_cost(
features, features_mask, labels, labels_mask,
speaker, start_flag, args.batch_size, raw_audio=raw_sequence)
cost_name = args.which_cost
cost.name = cost_name
if parrot.raw_output:
cost_raw.name = "sampleRNN_cost"
cg = ComputationGraph(cost)
model = Model(cost)
parameters = cg.parameters
step_rule = CompositeRule(
[StepClipping(10. * args.grad_clip), Adam(args.learning_rate)])
'timing_coeff': args.timing_coeff,
'encoder_type': saved_args.encoder_type,
'raw_output': saved_args.raw_output,
'name': 'parrot'
}
parrot = Parrot(**parrot_args)
features, features_mask, labels, labels_mask, speaker, start_flag, raw_audio = \
parrot.symbolic_input_variables()
cost, extra_updates, attention_vars, cost_raw = parrot.compute_cost(
features,
features_mask,
labels,
labels_mask,
speaker,
start_flag,
args.num_samples,
raw_audio=raw_audio)
model = Model(cost)
model.set_parameter_values(parameters)
print "Successfully loaded the parameters."
if args.sample_one_step:
gen_x, gen_k, gen_w, gen_pi, gen_phi, gen_pi_att = \
parrot.sample_using_input(data_tr, args.num_samples)
else:
gen_x, gen_k, gen_w, gen_pi, gen_phi, gen_pi_att = parrot.sample_model(
'attention_type': args.attention_type,
'attention_alignment': args.attention_alignment,
'encoder_type': args.encoder_type,
'weights_init': w_init,
'biases_init': b_init,
'name': 'parrot'
}
parrot = Parrot(**parrot_args)
parrot.initialize()
features, features_mask, labels, labels_mask, speaker, start_flag = \
parrot.symbolic_input_variables()
cost, extra_updates, attention_vars = parrot.compute_cost(
features, features_mask, labels, labels_mask, speaker, start_flag,
args.batch_size)
cost_name = args.which_cost
cost.name = cost_name
cg = ComputationGraph(cost)
model = Model(cost)
parameters = cg.parameters
step_rule = CompositeRule(
[StepClipping(10. * args.grad_clip),
Adam(args.learning_rate)])
algorithm = GradientDescent(cost=cost,