-
Notifications
You must be signed in to change notification settings - Fork 0
/
perplexity.py
232 lines (187 loc) · 9.48 KB
/
perplexity.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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
import os
import random
import argparse
import time
import tensorflow as tf
import numpy as np
import pandas as pd
from utils import load_data
class LSTMLM(tf.keras.Model):
def __init__(self, emb_dim, rnn_dim, vocab_size):
super().__init__()
self.embeddings = tf.keras.layers.Embedding(vocab_size, emb_dim, mask_zero=True)
self.rnn = tf.keras.layers.LSTM(rnn_dim, return_state=False, return_sequences=True,
kernel_initializer='lecun_normal', recurrent_initializer='lecun_normal')
self.vocab_prob = tf.keras.layers.Dense(vocab_size, activation='softmax')
@staticmethod
def reconstruction_loss(x, predictions):
# ignore padding
temp_mask = 1 - tf.keras.backend.cast_to_floatx(tf.keras.backend.equal(x, 0))
prob = tf.keras.backend.sparse_categorical_crossentropy(x, predictions) * temp_mask
res = tf.keras.backend.sum(prob, axis=-1)
return res
def call(self, x):
y = tf.keras.backend.concatenate([tf.constant(1, shape=(x.shape[0], 1)), x[:, :-1]], axis=-1)
embeddings = self.embeddings(y)
mask = self.embeddings.compute_mask(y)
rnn_output = self.rnn(embeddings, mask=mask)
predictions = self.vocab_prob(rnn_output)
rec_loss = self.reconstruction_loss(x, predictions)
return rec_loss
def train(self, optimizer, epochs, train_dataset, val_dataset, ckpt_man=None):
@tf.function
def train_step(x):
with tf.GradientTape() as tape:
loss = self(x)
loss = tf.keras.backend.mean(loss)
grads = tape.gradient(loss, self.weights)
optimizer.apply_gradients(zip(grads, self.weights))
return loss
@tf.function
def test_step(x):
loss = self(x)
return tf.keras.backend.mean(loss)
total_loss = 0
for step, x_batch_val in enumerate(val_dataset):
loss = test_step(x_batch_val)
total_loss = total_loss + loss
val_loss = total_loss / (step + 1)
print("loss:{:.4f}".format(val_loss))
# please refer to https://keras.io/guides/writing_a_training_loop_from_scratch/
step_count = 1
for epoch in range(1, epochs + 1):
print("Start of epoch {:d}".format(epoch))
start_time = time.time()
total_loss = 0
for step, x_batch_train in enumerate(train_dataset):
loss = train_step(x_batch_train)
total_loss = total_loss + loss
if step_count % 100 == 0:
print("step:{:d} train_loss:{:.4f}".format(step_count, loss))
step_count = step_count + 1
train_loss = total_loss / (step + 1)
print("train_loss:{:.4f}".format(train_loss))
total_loss = 0
for step, x_batch_val in enumerate(val_dataset):
loss = test_step(x_batch_val)
total_loss = total_loss + loss
val_loss = total_loss / (step + 1)
print("loss:{:.4f}".format(val_loss))
if ckpt_man is not None:
ckpt_man.save()
print("time taken:{:.2f}s".format(time.time() - start_time))
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='perplexity')
parser.add_argument('-s', '--seed', default=0, type=int, help='random seed')
parser.add_argument('-m', '--mpath', default='test', help='path of model')
args = parser.parse_args()
seed = args.seed
model_path = args.mpath
if not os.path.exists(os.path.join(os.path.join(os.getcwd(), 'model'), 'perplexity.txt')):
df = pd.DataFrame(columns=['Model', 'Forward', 'Reverse'])
df.to_csv(os.path.join(os.path.join(os.getcwd(), 'model'), 'perplexity.txt'), index=False, float_format='%.6f')
df = pd.read_csv(os.path.join(os.path.join(os.getcwd(), 'model'), 'perplexity.txt'))
if os.path.basename(model_path) in list(df['Model']):
print('Results already exists.')
exit()
else:
tf.random.set_seed(seed)
np.random.seed(seed)
random.seed(seed)
dic = {'Model': os.path.basename(model_path)}
with open(os.path.join(model_path, 'epoch_loss.txt'), 'r') as f:
s = f.readlines()[0]
if 'type' not in s:
model_type = 'VAE'
else:
model_type = s.split(',')[0].split()[-1]
if model_type == 'VAE':
s = s.split(',')
prior = s[0].split()[-1]
posterior = s[1].split()[-1]
emb_dim = int(s[2].split()[-1])
rnn_dim = int(s[3].split()[-1])
z_dim = int(s[4].split()[-1])
batch_size = int(s[5].split()[-1])
lr = float(s[7].split()[-1])
datapath = os.path.join(os.path.join(os.getcwd(), 'Dataset'), os.path.basename(s[-2].split()[-1]))
vocab_size = int(s[-1].split()[-1])
elif model_type == 'AE':
s = s.split(',')
emb_dim = int(s[1].split()[-1])
rnn_dim = int(s[2].split()[-1])
z_dim = int(s[3].split()[-1])
batch_size = int(s[4].split()[-1])
lr = float(s[6].split()[-1])
datapath = os.path.join(os.path.join(os.getcwd(), 'Dataset'), os.path.basename(s[-2].split()[-1]))
vocab_size = int(s[-1].split()[-1])
else:
s = s.split(',')
emb_dim = int(s[1].split()[-1])
rnn_dim = int(s[2].split()[-1])
z_dim = int(s[3].split()[-1])
batch_size = int(s[4].split()[-1])
lr = float(s[6].split()[-1])
datapath = os.path.join(os.path.join(os.getcwd(), 'Dataset'), os.path.basename(s[-2].split()[-1]))
vocab_size = int(s[-1].split()[-1])
word2index, index2word, train_dataset, val_dataset, test_dataset = load_data(batch_size, datapath,
is_train=True)
optimizer = tf.keras.optimizers.Adam(learning_rate=lr)
model = LSTMLM(emb_dim, rnn_dim, vocab_size)
lm_path = os.path.join(os.path.join(os.getcwd(), 'model'), os.path.basename(datapath) + '_lm')
if os.path.exists(lm_path):
ckpt = tf.train.Checkpoint(optimizer=optimizer, model=model)
ckpt.restore(tf.train.latest_checkpoint(lm_path)).expect_partial()
else:
ckpt_dir = lm_path
ckpt = tf.train.Checkpoint(optimizer=optimizer, model=model)
ckpt_man = tf.train.CheckpointManager(ckpt, directory=ckpt_dir, max_to_keep=1, checkpoint_name='ckpt')
model.train(optimizer, 10, train_dataset, val_dataset, ckpt_man)
total_ppl = 0
count = 0
for step, x_batch in enumerate(test_dataset):
sen_len = tf.math.count_nonzero(x_batch, axis=1)
batch_ppl = tf.keras.backend.exp(model(x_batch) / tf.keras.backend.cast_to_floatx(sen_len))
count += batch_ppl.shape[0]
total_ppl = total_ppl + tf.keras.backend.sum(batch_ppl)
df = pd.concat([df, pd.DataFrame(
{'Model': os.path.basename(datapath) + 'Real', 'Forward': float(total_ppl / count)}, index=[0])], ignore_index=True)
# test data
sentences = []
maxlen = 0
with open(os.path.join(model_path, 'generation.txt'), 'r') as f:
for sentence in f.readlines():
sentence = sentence.rstrip() + ' <eos>'
sentence = sentence.split()
for i in range(len(sentence)):
sentence[i] = word2index[sentence[i]]
if len(sentence) > maxlen:
maxlen = len(sentence)
sentences.append(sentence)
x_gen = tf.keras.preprocessing.sequence.pad_sequences(sentences, maxlen=maxlen, padding='post',
truncating='post')
gen_dataset = tf.data.Dataset.from_tensor_slices(x_gen)
gen_dataset = gen_dataset.batch(batch_size)
total_ppl = 0
count = 0
for step, x_batch in enumerate(gen_dataset):
sen_len = tf.math.count_nonzero(x_batch, axis=1)
batch_ppl = tf.keras.backend.exp(model(x_batch) / tf.keras.backend.cast_to_floatx(sen_len))
count += batch_ppl.shape[0]
total_ppl = total_ppl + tf.keras.backend.sum(batch_ppl)
dic['Forward'] = float(total_ppl / count)
gen_dataset = gen_dataset.shuffle(len(x_gen))
optimizer = tf.keras.optimizers.Adam(learning_rate=lr)
model = LSTMLM(emb_dim, rnn_dim, vocab_size)
model.train(optimizer, 10, gen_dataset, val_dataset)
total_ppl = 0
count = 0
for step, x_batch in enumerate(test_dataset):
sen_len = tf.math.count_nonzero(x_batch, axis=1)
batch_ppl = tf.keras.backend.exp(model(x_batch) / tf.keras.backend.cast_to_floatx(sen_len))
count += batch_ppl.shape[0]
total_ppl = total_ppl + tf.keras.backend.sum(batch_ppl)
dic['Reverse'] = float(total_ppl / count)
print(dic)
df = pd.concat([df, pd.DataFrame(dic, index=[0])], ignore_index=True)
df.to_csv(os.path.join(os.path.join(os.getcwd(), 'model'), 'perplexity.txt'), index=False, float_format='%.6f')