-
Notifications
You must be signed in to change notification settings - Fork 5
/
util.py
333 lines (299 loc) · 12.7 KB
/
util.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
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
from collections import defaultdict
import numpy as np
import math, ast, os, codecs
import cPickle as pickle
import json, sys, io
flatten = lambda l:[item for sublist in l for item in sublist]
recursive_flatten = lambda l:flatten([recursive_flatten(item) if isinstance(item, list) else [item] for item in l])
def normalize(x):
denom = sum(x)
return [i/denom for i in x]
def softmax(x):
e_x = np.exp(x - np.max(x))
out = e_x / e_x.sum()
return out
def weightedChoice(weights, objects, apply_softmax=False, alpha=None):
"""Return a random item from objects, with the weighting defined by weights
(which must sum to 1)."""
if apply_softmax: weights = softmax(weights)
if alpha: weights = normalize([w**alpha for w in weights])
cs = np.cumsum(weights) #An array of the weights, cumulatively summed.
idx = sum(cs < np.random.rand()) #Find the index of the first weight over a random value.
idx = min(idx, len(objects)-1)
return objects[idx]
def itersubclasses(cls, _seen=None):
if not isinstance(cls, type):
raise TypeError('itersubclasses must be called with '
'new-style classes, not %.100r' % cls)
if _seen is None: _seen = set()
try:
subs = cls.__subclasses__()
except TypeError: # fails only when cls is type
subs = cls.__subclasses__(cls)
for sub in subs:
if sub not in _seen:
_seen.add(sub)
yield sub
for sub in itersubclasses(sub, _seen):
yield sub
def get_no_padding_indices(data, max_size):
ans = []
current_start = 0
current_count = 0
for i in range(len(data)):
if current_count >= max_size or len(data[current_start]) != len(data[i]):
ans.append((current_start, i))
current_count = 0
current_start = i
current_count += 1
ans.append((current_start, i+1))
return ans
class key_dependent_dict(defaultdict):
def __init__(self,f_of_x):
super(key_dependent_dict, self).__init__(None) # base class doesn't get a factory
self.f_of_x = f_of_x # save f(x)
def __missing__(self, key): # called when a default needed
ret = self.f_of_x(key) # calculate default value
self[key] = ret # and install it in the dict
return ret
class Token(object):
def __init__(self, i, s, count=1):
self.i = i
self.s = s
self.count = count
self.original_s = s
def __eq__(self, other):
return self.i == other or self.s == other or \
(isinstance(other, Token) and self.i == other.i and self.s == other.s)
def __str__(self): return unicode(self.s)
def __repr__(self): return str((self.s, self.i))
def __hash__(self): return self.i
@staticmethod
def not_found(): raise Exception("token not found")
class Vocab(object):
def __init__(self):
self.tokens = set([])
self.strings = set([])
self.s2t = defaultdict(Token.not_found)
self.i2t = defaultdict(Token.not_found)
self.unk = None
self.START_TOK = None
self.END_TOK = None
self.delimiter = u''
@property
def size(self):
return len(self.strings)
def add(self, thing):
if isinstance(thing, Token): self.add_token(thing)
else: self.add_string(thing)
def add_string(self, string):
if string in self.strings:
self[string].count += 1
return self[string]
i = len(self.tokens)
s = string
t = Token(i, s)
self.i2t[i] = t
self.s2t[s] = t
self.tokens.add(t)
self.strings.add(s)
return t
def add_token(self, tok):
self.i2t[tok.i] = tok
self.s2t[tok.s] = tok
self.tokens.add(tok)
self.strings.add(tok.s)
return tok
def __getitem__(self, key):
if isinstance(key, int): return self.i2t[key]
elif isinstance(key, Token): return key
else: return self.s2t[key]
def unkify(self, string):
tok = self.unk
tok.original = string
return tok
def add_unk(self, thresh=0, unk_string='<UNK>'):
if unk_string in self.s2t.keys(): raise Exception("tried to add an UNK token that already existed")
if self.unk is not None: raise Exception("already added an UNK token")
strings = [unk_string]
for token in self.tokens:
if token.count >= thresh: strings.append(token.s)
if self.START_TOK is not None and self.START_TOK not in strings: strings.append(self.START_TOK.s)
if self.END_TOK is not None and self.END_TOK not in strings: strings.append(self.END_TOK.s)
self.tokens = set([])
self.strings = set([])
self.i2t = key_dependent_dict(lambda s:self.unkify(s))
self.s2t = key_dependent_dict(lambda s:self.unkify(s))
for string in strings:
self.add_string(string)
self.unk = self.s2t[unk_string]
if self.START_TOK is not None: self.START_TOK = self.s2t[self.START_TOK.s]
if self.END_TOK is not None: self.END_TOK = self.s2t[self.END_TOK.s]
def pp(self, seq, delimiter=None):
if delimiter is None: delimiter = self.delimiter
return delimiter.join([unicode(self[item].s) for item in seq])
def hpp(self, seq, delimiter=u''):
if delimiter is None: delimiter = self.delimiter
if isinstance(seq, int): return self.i2t[seq]
else: return "["+delimiter.join([self.hpp(thing) for thing in seq])+"]"
def save(self, filename):
info_dict = {
"tokens":self.tokens,
"strings":self.strings,
"s2t":dict(self.s2t),
"i2t":dict(self.i2t),
"unk":self.unk,
"delimiter":self.delimiter,
"START_TOK":self.START_TOK,
"END_TOK":self.END_TOK
}
with open(filename, "w") as f: pickle.dump(info_dict, f)
@classmethod
def load(cls, filename):
with open(filename, "r") as f:
info_dict = pickle.load(f)
v = Vocab()
v.tokens = info_dict["tokens"]
v.strings = info_dict["strings"]
v.unk = info_dict["unk"]
v.delimiter = info_dict["delimiter"]
defaultf = (lambda :v.unk) if (v.unk is not None) else Token.not_found
v.s2t = defaultdict(defaultf, info_dict["s2t"])
v.i2t = defaultdict(defaultf, info_dict["i2t"])
v.START_TOK = info_dict["START_TOK"]
v.END_TOK = info_dict["END_TOK"]
return v
@classmethod
def load_from_corpus(cls, reader, remake=False, src_or_tgt="src"):
vocab_fname = reader.fname+".vocab."+reader.mode+"."+src_or_tgt
if not remake and os.path.isfile(vocab_fname):
return Vocab.load(vocab_fname)
else:
v = Vocab()
count = 0
for item in reader:
if reader.seq2seq:
if src_or_tgt == "src": toklist = item[0]
if src_or_tgt == "tgt": toklist = item[1]
else:
toklist = item
for token in toklist:
v.add(token)
count += 1
if count % 100 == 0:
print "...", count,
sys.stdout.flush()
v.delimiter = reader.delimiter
v.START_TOK = v[reader.begin] if reader.begin is not None else None
v.END_TOK = v[reader.end] if reader.end is not None else None
print "saving vocab of size", v.size
v.save(vocab_fname)
return v
#### reader classes
class CorpusReaderTemplate(object):
names = {"template",}
def get_reader(name):
for c in itersubclasses(CorpusReaderTemplate):
if name in c.names: return c
raise Exception("no reader found with name: " + name)
class GenericLineCorpusReader(CorpusReaderTemplate):
names = {"generic_char", "generic_word", "tokenize_word"}
def __init__(self, fname, begin=None, end=None, mode="generic_char"):
self.fname = fname
self.mode = mode
self.begin = begin
self.end = end
self.seq2seq = False
self.delimiter = '' if mode == "generic_char" else ' '
if mode == "tokenize_word":
from pattern.en import tokenize
self.tokenize = tokenize
def __iter__(self):
if os.path.isdir(self.fname):
filenames = [os.path.join(self.fname,f) for f in os.listdir(self.fname)]
else:
filenames = [self.fname]
for filename in filenames:
# with io.open(filename, encoding='utf-8') as f:
with open(filename) as f:
doc = f.read()
if self.mode == "generic_char":
for line in doc.split("\n"):
if not line: continue
yield [self.begin] + list(line) + [self.end]
elif self.mode == "generic_word":
for line in doc.split("\n"):
if not line: continue
yield [self.begin] + [word for word in line.split(" ") if word] + [self.end]
elif self.mode == "tokenize_word":
for line in doc.split("\n"):
if not line: continue
yield [self.begin] + [word for word in ' '.join(self.tokenize(line)).split(" ") if word] + [self.end]
class PTBCorpusReader(CorpusReaderTemplate):
names = {"ptb", "ptb_bracketed", "ptb_stripped", "ptb_char_stripped", "ptb_nopos"}
def __init__(self, fname, begin=None, end=None, mode="ptb_bracketed"):
self.fname = fname
self.mode = mode
self.begin = begin
self.end = end
self.seq2seq = False
if "char" in mode:
self.delimiter = '\032'
else:
self.delimiter = ' '
def __iter__(self):
if os.path.isdir(self.fname):
filenames = [os.path.join(self.fname,f) for f in os.listdir(self.fname)]
else:
filenames = [self.fname]
for filename in filenames:
# with io.open(filename, encoding='utf-8') as f:
with open(filename) as f:
doc = f.read()
for oline in doc.split("\n"):
line = oline.strip()
if not line: continue
line = line.replace(")", " )")
line = line.split(" ")
if self.mode == "ptb":
yield [self.begin] + [item for item in line if item] + [self.end]
else:
line = [line[i] for i in range(len(line)) if not ((line[i][0] == "(" and i+2 < len(line) and line[i+2][0] == ")")
or (line[i][0] == ")" and i-2 >= 0 and line[i-2][0] == "("))]
if self.mode == "ptb_nopos":
ans = [item for item in line if item]
yield [self.begin] + ans + [self.end]
elif self.mode == "ptb_stripped":
yield [self.begin] + [item for item in line if item and item[0] not in {"(", ")"}] + [self.end]
elif self.mode == "ptb_char_stripped":
yield [self.begin] + list(' '.join([item for item in line if item and item[0] not in {"(", ")"}])) + [self.end]
elif self.mode == "ptb_bracketed":
ans = [self.begin]
bracket_stack = []
for item in line:
if item[0] == "(":
ans.append(item)
bracket_stack.append(item[1:])
elif item[0] == ")":
ans.append(item+bracket_stack.pop(-1))
else:
ans.append(item)
ans.append(self.end)
yield ans
elif self.mode == "ptb_brackets_only":
ans = [self.begin]
bracket_stack = []
for item in line:
if item[0] == "(":
ans.append(item)
bracket_stack.append(item[1:])
elif item[0] == ")":
ans.append(item+bracket_stack.pop(-1))
else:
ans.append("<WORD>")
ans.append(self.end)
yield ans
elif self.mode == "raw":
yield oline
else:
raise Exception("bad mode:" + self.mode)