break
kaldiIn.close()
input_shape_train = conv_configs[0]['input_shape']
input_shape_1 = (input_shape_train[1], input_shape_train[2], input_shape_train[3])
num_utt = len(feat_mats_np)
feat_mats = []
for i in xrange(num_utt):
feat_mats.append(numpy.reshape(feat_mats_np[i], (feat_rows[i],) + input_shape_1))
rng = numpy.random.RandomState(123)
theano_rng = RandomStreams(rng.randint(2 ** 30))
cnn = CNN_Forward(numpy_rng = rng, theano_rng=theano_rng,
conv_layer_configs = conv_configs, use_fast = use_fast)
_file2cnn(cnn.conv_layers, filename=conv_net_file)
out_function = cnn.build_out_function(feat_mats)
log('> ... processing the data')
kaldiOut = KaldiWriteOut(output_scp,output_ark)
kaldiOut.open()
for i in xrange(num_utt):
feat_out = out_function(feat_mats[i])
kaldiOut.write(uttIDs[i], feat_out)
kaldiOut.close()
end_time = time.clock()
print >> sys.stderr, ('The code for file ' +
os.path.split(__file__)[1] +
' ran for %.2fm' % ((end_time - start_time) / 60.))
numpy_rng = numpy.random.RandomState(89677)
theano_rng = RandomStreams(numpy_rng.randint(2 ** 30))
log('> ... building the model')
# construct the cnn architecture
cnn = CNN_SAT(numpy_rng=numpy_rng, theano_rng = theano_rng,
batch_size = batch_size, n_outs=n_outs,
conv_layer_configs = conv_layer_configs,
hidden_layers_sizes = hidden_layers_sizes,
ivec_layers_sizes = ivec_layers_sizes,
conv_activation = conv_activation,
full_activation = full_activation,
use_fast = use_fast, update_part = update_part, ivec_dim = ivec_dim)
if arguments.has_key('conv_input_file'):
_file2cnn(cnn.conv_layers, filename=arguments['conv_input_file'])
if arguments.has_key('full_input_file'):
_file2nnet(cnn.full_layers, filename = arguments['full_input_file'])
if arguments.has_key('ivec_input_file'):
_file2nnet(cnn.ivec_layers, set_layer_num = len(ivec_layers_sizes) + 1, filename = arguments['ivec_input_file'], withfinal=False)
# get the training, validation and testing function for the model
log('> ... getting the finetuning functions')
train_fn, valid_fn = cnn.build_finetune_functions(
(train_x, train_y), (valid_x, valid_y),
batch_size=batch_size)
log('> ... finetunning the model')
start_time = time.clock()
while (lrate.get_rate() != 0):
theano_rng = RandomStreams(numpy_rng.randint(2 ** 30))
log('> ... building the model')
# construct the cnn architecture
cnn = CNN(numpy_rng=numpy_rng, theano_rng = theano_rng,
batch_size = batch_size, n_outs=n_outs,
conv_layer_configs = conv_layer_configs,
hidden_layers_sizes = hidden_layers_sizes,
conv_activation = conv_activation,
full_activation = full_activation,
use_fast = use_fast, update_layers = update_layers)
total_layer_number = len(cnn.layers)
if full_ptr_layer_number > 0:
_file2nnet(cnn.layers[len(conv_layer_configs):total_layer_number], set_layer_num = full_ptr_layer_number, filename = full_ptr_file, withfinal=False)
if conv_ptr_layer_number > 0:
_file2cnn(cnn.layers[0:len(conv_layer_configs)], filename=conv_ptr_file)
# get the training, validation and testing function for the model
log('> ... getting the finetuning functions')
train_fn, valid_fn = cnn.build_finetune_functions(
(train_x, train_y), (valid_x, valid_y),
batch_size=batch_size)
log('> ... finetunning the model')
start_time = time.clock()
while (lrate.get_rate() != 0):
train_error = []
while (not train_sets.is_finish()):
train_sets.load_next_partition(train_xy)
for batch_index in xrange(train_sets.cur_frame_num / batch_size): # loop over mini-batches
train_error.append(train_fn(index=batch_index, learning_rate = lrate.get_rate(), momentum = momentum))
