-
Notifications
You must be signed in to change notification settings - Fork 9
/
LasagneCLM.py
132 lines (94 loc) · 4.15 KB
/
LasagneCLM.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
import numpy as np
import os
import theano
import theano.tensor as T
import lasagne
import pdb
# from ctc import CTCLayer
# from TIMIT_utils import create_mask
import pickle
import itertools
# import matplotlib.pyplot as plt
####### SETTING DEFAULT HYPER-PARAMETER VALUES #######
#Input representation size
INPUT_SIZE = 29
#Hidden layer hyper-parameters
N_HIDDEN = 100
HIDDEN_NONLINEARITY = 'rectify'
#Learning rate
LEARNING_RATE = 0.001
#Number of training sequences (here sentences or examples) in each batch
N_BATCH = 100
#Gradient clipping
GRAD_CLIP = 100
#How often we check the output
EPOCH_SIZE = 100
#Number of epochs to train the system on
NUM_EPOCHS = 100
#Data file name
DATA_FILE_NAME = os.path.join(os.getcwd(),'TIMIT_data_prepared_for_CLM.pkl')
# pdb.set_trace()
# def main(num_epochs = NUM_EPOCHS):
print("Building the network...")
l_in = lasagne.layers.InputLayer(shape = (None, None, INPUT_SIZE)) #One-hot represenntation of character indices
l_mask = lasagne.layers.InputLayer(shape = (None, None))
l_recurrent = lasagne.layers.RecurrentLayer(incoming = l_in, num_units=N_HIDDEN, mask_input = l_mask, learn_init=True)
Recurrent_output=lasagne.layers.get_output(l_recurrent)
n_batch, n_time_steps, n_features = l_in.input_var.shape
l_reshape = lasagne.layers.ReshapeLayer(l_recurrent, (-1, N_HIDDEN))
Reshape_output = lasagne.layers.get_output(l_reshape)
l_dense = lasagne.layers.DenseLayer(l_reshape, num_units=INPUT_SIZE, nonlinearity = lasagne.nonlinearities.softmax)
# l_out = lasagne.layers.ReshapeLayer(l_dense, (n_batch, n_time_steps, n_features))
#Training the network
target_values = T.ivector('target_output') #A vector of character indices linearized over the sequences and batches
target_mask = T.vector('target_mask',dtype=theano.config.floatX)
#Getting the expression for the output (we extract it from the softmax layer to make sure it is compatible with the cross entropy loss function of theano)
network_output = lasagne.layers.get_output(l_dense)
#Calculating the cross-entropy loss
# cost = T.mean(lasagne.objectives.categorical_crossentropy(network_output, target_values))
cost_pointwise = lasagne.objectives.categorical_crossentropy(network_output, target_values)
cost = (cost_pointwise*target_mask).sum()
# cost.reshape([n_batch, n_time_steps, n_features])
all_params = lasagne.layers.get_all_params(l_dense)
updates = lasagne.updates.adam(cost, all_params)
train = theano.function(inputs=[l_in.input_var, target_values, l_mask.input_var, target_mask], outputs=[cost], updates=updates)
compute_cost = theano.function( inputs=[l_in.input_var, target_values, l_mask.input_var, target_mask], outputs=[cost], )
# pdb.set_trace()
#################################################################################
#Load data
with open(DATA_FILE_NAME,'rb') as f:
data = pickle.load(f)
x = data['x']
mask = data['mask']
y = data['y']
y_merged = y.reshape([-1])
mask_merged = mask.reshape([-1])
#Preparing one single batch of inputs
print('Input shape: {}'
'Total output len: {}'.format(x.shape, len(y_merged)))
Recurrent_output_value=Recurrent_output.eval({l_in.input_var: x, l_mask.input_var: mask})
network_output_value=network_output.eval({l_in.input_var: x, l_mask.input_var: mask})
cost_values_pointwise = cost_pointwise.eval({l_in.input_var: x, l_mask.input_var: mask, target_values: y_merged})
cost_value = cost.eval({l_in.input_var: x, l_mask.input_var: mask, target_values: y_merged, target_mask: mask_merged})
print("Recurrent_output"
"Shape: {}"
"datatype: {}".format(Recurrent_output_value.shape, type(Recurrent_output_value)))
print("network output"
"Shape: {}"
"datatype: {}".format(network_output_value.shape, type(network_output_value)))
print("Pointwise Cost: {}"
"Cost: {}".format(cost_values_pointwise.shape, cost_value))
cost_vector = []
for epoch in range(NUM_EPOCHS):
shuffle_order = np.random.permutation(x.shape[0])
x = x[shuffle_order, :]
y = y[shuffle_order, :]
y_merged = y.reshape([-1])
mask = mask[shuffle_order, :]
mask_merged = mask.reshape([-1])
cost = train(x,y_merged,mask,mask_merged)
print("Epoch: {}"
"\tcost = {}".format(epoch,cost) )
cost_vector.append(cost)
# plt.plot(np.arange(NUM_EPOCHS),cost_vector)
# plt.show()