forked from anirudh9119/LM_GANS
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Descriminator.py
143 lines (106 loc) · 4.68 KB
/
Descriminator.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
133
134
135
136
137
138
139
140
141
142
import theano
import theano.tensor as T
import numpy as np
import random
from lasagne.layers import DenseLayer, LSTMLayer
from pad_list import pad_list
import lasagne
from data_iterator import TextIterator
import cPickle as pkl
'''
-Build a discriminator.
-Each time we train, use 1 for "real" and 0 for "sample".
-In later uses, we'll need to define a different transformation for sampling from the generator-RNN which is differentiable.
-Takes input matrix of integers.
-For each time step, index into word matrix using saved indices.
'''
class discriminator:
def __init__(self, number_words, num_hidden, seq_length, mb_size):
self.mb_size = mb_size
x = T.imatrix()
target = T.ivector()
word_embeddings = theano.shared(np.random.normal(size = ((number_words, 1, num_hidden))).astype('float32'))
feature_lst = []
for i in range(0, seq_length):
feature = word_embeddings[x[:,i]]
feature_lst.append(feature)
features = T.concatenate(feature_lst, 1)
#example x sequence_position x feature
#inp = InputLayer(shape = (seq_length, mb_size, num_hidden), input_var = features)
l_lstm_1 = LSTMLayer((seq_length, mb_size, num_hidden), num_units = num_hidden, nonlinearity = lasagne.nonlinearities.tanh)
l_lstm_2 = LSTMLayer((seq_length, mb_size, num_hidden), num_units = num_hidden, nonlinearity = lasagne.nonlinearities.tanh)
#minibatch x sequence x feature
final_out = T.mean(l_lstm_2.get_output_for([l_lstm_1.get_output_for([features])]), axis = 1)
#final_out = T.mean(features, axis = 1)
h_out = DenseLayer((mb_size, num_hidden), num_units = 1, nonlinearity=None)
h_out_value = h_out.get_output_for(final_out)
classification = T.nnet.sigmoid(h_out_value)
self.loss = T.mean(T.nnet.binary_crossentropy(output = classification.flatten(), target = target))
self.params = lasagne.layers.get_all_params(h_out,trainable=True) + [word_embeddings] + lasagne.layers.get_all_params(l_lstm_1, trainable = True) + lasagne.layers.get_all_params(l_lstm_2, trainable = True)
updates = lasagne.updates.adam(self.loss, self.params)
self.train_func = theano.function(inputs = [x, target], outputs = {'l' : self.loss, 'c' : classification}, updates = updates)
self.evaluate_func = theano.function(inputs = [x], outputs = {'c' : classification})
def train_real(self, x):
return self.train_func(x, [1] * self.mb_size)
def train_fake(self, x):
return self.train_func(x, [0] * self.mb_size)
def evaluate(self, x):
return self.evaluate_func(x)
if __name__ == "__main__":
seq_length = 30
dictionary='/data/lisatmp4/anirudhg/wiki.tok.txt.gz.pkl'
valid_dataset='/data/lisatmp4/anirudhg/temp.en.tok'
gen_dataset = 'temp_workfile'
batch_size = 1
n_words = 30000
maxlen = 30
# load dictionary
with open(dictionary, 'rb') as f:
worddicts = pkl.load(f)
# invert dictionary
worddicts_r = dict()
for kk, vv in worddicts.iteritems():
worddicts_r[vv] = kk
print 'Loading data'
actual_sentences = TextIterator(valid_dataset,
dictionary,
n_words_source = n_words,
batch_size = batch_size,
maxlen=maxlen)
gen_sentences = TextIterator(gen_dataset,
dictionary,
n_words_source = n_words,
batch_size = batch_size,
maxlen=maxlen)
orig_sen = []
for x in actual_sentences:
orig_sen.append(x[0])
#orig_sen.append(numpy.zeros(30).tolist())
orig_s = pad_list(orig_sen)
print orig_s.shape
gen_sen = []
for x in gen_sentences:
gen_sen.append(x[0])
#gen_sen.append(numpy.zeros(30).tolist())
print len(gen_sen)
print len(orig_sen)
gen_s = pad_list(gen_sen)
print gen_s.shape
#5000 gen
#2144 orig
orig_s = np.loadz('orig_s.npz')
gen_s = np.loadz('gen_s.npz')
print "compiling"
d = discriminator(number_words = 30000, num_hidden = 400, seq_length = seq_length, mb_size = 64)
print "training started"
for i in range(0,20):
u = random.uniform(0,1)
indexGen = random.randint(0, 200 / 64)
indexOrig = random.randint(0, 200 / 64)
print orig_s[indexOrig * 64 : (indexOrig + 1) * 64].shape
if u < 0.5:
d.train_real(orig_s[0 : 64].astype('int32'))
else:
d.train_fake(gen_s[0 : 64].astype('int32'))
# print "real", d.evaluate(orig_s[100:4000 + 64].astype('int32'))
# print "fake", d.evaluate(gen_s[1500:1500 + 64].astype('int32'))