/
myutil.py
295 lines (255 loc) · 10.1 KB
/
myutil.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
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
import torch
import torch.nn as nn
from torch.autograd import Variable
import numpy as np
import pickle
import logging
import os
logger = logging.getLogger()
class DialogueBatches(object):
def __init__(self, dirname, batch_size, src_max_len, tgt_max_len, vocab_inv, config, his_len = 2, lower = False, sw = '<eou>'):
logger.info('initing DialogueBatches lower: %s his_len: %s filter_turn: %s skip_unk: %s', str(lower), str(his_len), str(config['filter_turn']), str(config['skip_unk']))
self.dirname = dirname
self.src_max_len = src_max_len
self.tgt_max_len = tgt_max_len
self.batch_size = batch_size
self.vocab_inv = vocab_inv
self.lower = lower
self.his_len = his_len
self.filter_turn = config['filter_turn']
self.skip_unk = config['skip_unk']
@staticmethod
def init_inf(dirname, batch_size, src_max_len, tgt_max_len, vocab_inv, config, his_len = 2, lower = False, sw = '<eou>', name = ''):
iter_count = 0
while True:
batches = DialogueBatches(dirname, batch_size, src_max_len, tgt_max_len, vocab_inv, config, his_len = his_len, lower = lower, sw = sw)
for a, b, c, d in batches:
yield a, b, c, d
iter_count = iter_count + 1
logger.info('data sweep time %d %s', iter_count, name)
def arrange_mb(self, res):
vocab_inv = self.vocab_inv
len_src = 0
len_tgt = 0
src_lis = []
tgt_lis = []
tgt_lens = []
for i in range(self.batch_size):
src = []
his = res[i]['his']
for h in his:
src.extend(h)
src.append('<eou>')
if len(src) > self.src_max_len:
src = src[len(src)-self.src_max_len:]
len_src = max(len_src, len(src))
src_lis.append(src)
tgt = res[i]['tgt']
tgt = ['<s>'] + tgt + ['</s>']
if len(tgt) > self.tgt_max_len:
tgt = tgt[:self.tgt_max_len]
tgt_lens.append(len(tgt) - 1)
len_tgt = max(len_tgt, len(tgt))
tgt_lis.append(tgt)
mb_src = torch.LongTensor(self.batch_size, len_src)
mb_tgt = torch.LongTensor(self.batch_size, len_tgt)
for i in range(self.batch_size):
src = src_lis[i]
tgt = tgt_lis[i]
src = ['<pad>'] * (len_src - len(src)) + src
tgt = tgt + ['<pad>'] * (len_tgt - len(tgt))
#print 'src:', src
#print 'tgt:', tgt
for j in range(len_src):
mb_src[i][j] = vocab_inv[src[j]] if src[j] in vocab_inv else vocab_inv['<unk>']
for j in range(len_tgt):
mb_tgt[i][j] = vocab_inv[tgt[j]] if tgt[j] in vocab_inv else vocab_inv['<unk>']
return mb_src, mb_tgt, tgt_lens, src_lis, tgt_lis
def __iter__(self):
vocab_inv = self.vocab_inv
if type(self.dirname) != type([]):
fns = os.listdir(self.dirname)
fns = [os.path.join(self.dirname, fn) for fn in fns]
else:
fns = self.dirname
res = []
for fname in fns:
for line in open(fname):
his = []
lines = line.split('<eou>')
if len(lines) <= 1: #we only consider the dialogues of more than two sentences
continue
co = 0
for l in lines:
co = co + 1
l = l.split()
if self.skip_unk == True:
s_l = [w for w in l if w in self.vocab_inv]
if len(s_l) < len(l):
print s_l, l
l = s_l
if len(l) <= 0:
continue
if len(his) > 0:
if co % self.filter_turn == 0:
res.append({'his':list(his), 'tgt':list(l)})
if len(res) == self.batch_size:
yield self.arrange_mb(res)
res = []
his.append(l)
if len(his) > self.his_len:
his = his[1:]
class MyBatchSentences_v2(object):
def __init__(self, dirname, batch_size, max_len, vocab_inv, lower = True, do_sort = True):
self.dirname = dirname
self.max_len = max_len
self.batch_size = batch_size
self.vocab_inv = vocab_inv
self.lower = lower
self.do_sort = do_sort
def __iter__(self):
ss_now = []
lens = []
vocab_inv = self.vocab_inv
if type(self.dirname) != type([]):
fns = os.listdir(self.dirname)
self.dirname = [os.path.join(self.dirname, fn) for fn in fns]
for fname in self.dirname:
for line in open(fname):
if self.lower == True:
l = line.strip().lower().split()
else:
l = line.strip().split()
if len(l) == 0:
continue
if l[-1] == '</s>' or l[-1] == '<eos>':
l = l[:-1]
l.append('</s>')
if len(l) > self.max_len - 1:
l = l[:self.max_len]
if l[0] == '<s>':
l = l[1:]
lens.append(len(l)) #len does not count <s>
l = ['<s>'] + l
ss_now.append(l)
if len(ss_now) == self.batch_size:
if self.do_sort == True:
ss_now, lens = length_sort(ss_now, lens)
ss_idx = [[(vocab_inv[w] if (w in vocab_inv) else vocab_inv['<unk>']) for w in l] for l in ss_now]
ss_idx = [(l + [0] * (max(lens) + 1 - len(l))) for l in ss_idx]
ss_idx = torch.LongTensor(ss_idx).cuda()
yield ss_idx, ss_now, lens
ss_now = []
lens = []
class MyStatDic(object):
def __init__(self):
self.dd = {}
def append_dict(self, new_d, keys = None):
dd = self.dd
if keys == None:
keys = new_d.keys()
for k in keys:
if not k in dd: dd[k] = []
dd[k].append(new_d[k])
def log_mean(self, keys = None, last_num = 0, log_pre = ''):
ss = ''; dd = self.dd;
if keys == None: keys = dd.keys()
for k in keys:
ss = ss + k + ': ' + str(np.mean(dd[k][-last_num:])) + ' '
logger.info(log_pre + ' (last_num %d ) ' + ss, last_num)
def force_onehot(sv):
pl = torch.zeros(sv.size()).cuda()
idx = torch.max(sv, dim = 2)[1]
for i1 in range(idx.size(0)):
for i2 in range(idx.size(1)):
pl[i1, i2, idx[i1, i2].item()] = 1
return pl
def clean_sen(sen):
if len(sen) == 0:
return sen
for i in range(len(sen)):
if sen[i] == '</s>':
sen = sen[:i]
break
while len(sen) > 0 and (sen[0] == '<pad>' or sen[0] == '<s>'):
sen = sen[1:]
while len(sen) > 0 and (sen[-1] == '<pad>' or sen[-1] == '</s>'):
sen = sen[:-1]
return sen
def length_sort(items, lengths, descending=True):
"""In order to use pytorch variable length sequence package"""
items = list(zip(items, lengths))
items.sort(key=lambda x: x[1], reverse=True)
items, lengths = zip(*items)
return list(items), list(lengths)
def getVocab(fn, lower = False):
logger.info('learning vocab from %s, lower: %s', fn, str(lower))
vocab = ['</s>', '<unk>', '<s>', '<pad>', '<eou>']
vocab_inv = {'</s>':0, '<unk>':1, '<s>':2, '<pad>':3, '<eou>':4}
for line in open(fn):
if lower == True:
l = line.strip().lower().split()
else:
l = line.strip().split()
for w in l:
if not w in vocab_inv:
vocab.append(w)
vocab_inv[w] = len(vocab) - 1
return vocab, vocab_inv
def mask_gen(lengths, ty = 'Byte'):
max_len = max(lengths)
size = len(lengths)
if ty == 'Byte':
mask = torch.ByteTensor(size, max_len).zero_()
elif ty == 'Float':
mask = torch.FloatTensor(size, max_len).zero_()
for i in range(size):
mask[i][:lengths[i]].fill_(1)
return mask
def setLogger(logger, LOG_FN):
logger.handlers = []
fileHandler = logging.FileHandler(LOG_FN, mode = 'w') #could also be 'a'
logFormatter = logging.Formatter("%(asctime)s [%(funcName)-15s] [%(levelname)-5.5s] %(message)s")
fileHandler.setFormatter(logFormatter)
logger.addHandler(fileHandler)
consoleHandler = logging.StreamHandler()
consoleHandler.setFormatter(logFormatter)
logger.addHandler(consoleHandler)
def check_memory():
# return the memory usage in MB
import psutil
process = psutil.Process(os.getpid())
mem = process.memory_info()[0] / float(2 ** 20)
#logger.info('current memory : %d MB', mem)
if mem > 7 * 1024: #large than 8G
logger.info('too much memory, forcing shutdown...')
sys.exit(1)
return mem
def idx2onehot(idx, vocab_size):
assert(len(idx.size()) == 2)
#idx should be a integer tensor
l = idx.size(0) * idx.size(1)
oh = torch.zeros(l, vocab_size).cuda()
oh[range(l), idx.contiguous().view(-1).cpu()] = 1
oh = oh.view(idx.size(0), idx.size(1), vocab_size).cuda()
return oh
def init_lstm_hidden(bsz, hz, layer_num = 1):
zeros1 = Variable(torch.zeros(layer_num, bsz, hz)).cuda()
zeros2 = Variable(torch.zeros(layer_num, bsz, hz)).cuda()
return zeros1, zeros2
def add_log_fn(fn, log_fn, save_dir):
logger.info('attaching log_fn %s to %s', log_fn, fn)
ff = open(fn, 'a')
ff.write('{} \t {}\n'.format(log_fn, save_dir))
ff.close()
def countSenAcc(target, dec, vocab_inv):
#accepts two lists of lists
c = 0
eos_id = vocab_inv['</s>']
for l in range(len(target)):
for j in range(len(target[l])):
if target[l][j] == dec[l][j]:
c = c + 1
if target[l][j] == eos_id:
break
return c