-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
339 lines (293 loc) · 15.8 KB
/
main.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
334
335
336
337
338
339
from corpus_sizes.global_params import *
from corpus_sizes.corpus_processing import corpus_processing
import scipy
import gensim.downloader as api
from gensim.models import Word2Vec
from gensim.models import FastText
from gensim.models import KeyedVectors
from web.embedding import Embedding
from web.datasets.similarity import fetch_MEN, fetch_MTurk, fetch_RG65, fetch_RW, fetch_SimLex999, fetch_TR9856, fetch_WS353
from web.datasets.analogy import fetch_google_analogy, fetch_msr_analogy, fetch_wordrep
from web.evaluate import evaluate_similarity
from analogy_solver import evaluate_analogy
from collections import Counter
from operator import itemgetter
import numpy as np
import pandas as pd
import pickle
import os
import datetime
def get_dataset(dataset_name):
if dataset_name == "WS353":
return fetch_WS353("similarity")
elif dataset_name == "MEN":
return fetch_MEN("all")
elif dataset_name == "SimLex-999":
return fetch_SimLex999()
elif dataset_name == "MTurk":
return fetch_MTurk()
elif dataset_name == "WS353":
return fetch_WS353('all')
elif dataset_name == "RG65":
return fetch_RG65()
elif dataset_name == "RW":
return fetch_RW()
elif dataset_name == "TR9856":
return fetch_TR9856()
elif dataset_name == "MSR":
return fetch_msr_analogy()
elif dataset_name == "Google":
return fetch_google_analogy()
else:
raise Exception("{}: dataset not supported".format(dataset_name))
def single_word_bins(sentences, vocabulary):
c = Counter(word for sentence in sentences for word in set(sentence))
word_freqs = [(word, c[word]) for word in vocabulary if c[word] > 0]
out_of_vocabulary = [(word, c[word]) for word in vocabulary if c[word] == 0]
word_freqs.sort(key = itemgetter(1, 0))
words = [word for (word, freq) in word_freqs]
data_len = len(word_freqs)
split = data_len // 3
high = words[2 * split:]
high_bounds = (c[high[0]], c[high[-1]])
mid = words[split:2 * split]
mid_bounds = (c[mid[0]], c[mid[-1]])
low = words[:split]
low_bounds = (c[low[0]], c[low[-1]])
return high, mid, low, high_bounds, mid_bounds, low_bounds, out_of_vocabulary
def pair_bins(low, mid, high, dataset):
pair_similarity ={}
low_pairs = []
mid_pairs = []
high_pairs = []
mixed_pairs = []
low_similarity = []
mid_similarity = []
high_similarity = []
mixed_similarity = []
for pair, similarity in zip(dataset.X, dataset.y):
if pair[0] in low and pair[1] in low:
low_pairs.append(pair)
low_similarity.append(similarity)
continue
if pair[0] in mid and pair[1] in mid:
mid_pairs.append(pair)
mid_similarity.append(similarity)
continue
if pair[0] in high and pair[1] in high:
high_pairs.append(pair)
high_similarity.append(similarity)
continue
mixed_pairs.append(pair)
mixed_similarity.append(similarity)
pair_similarity['low'] = (low_pairs, low_similarity)
pair_similarity['mid'] = (mid_pairs, mid_similarity)
pair_similarity['high'] = (high_pairs, high_similarity)
pair_similarity['mixed'] = (mixed_pairs, mixed_similarity)
return pair_similarity
def save_results(embedding_type, gen_similarity, low_similarity, mid_similarity, high_similarity, mixed_similarity,
out_of_vocab, low_count, mid_count, high_count, mixed_count, low_bounds, mid_bounds, high_bounds,
dim, ww, dataset_name, corpus_size):
method = "SG" if SKIP_GRAMS else "CBOW"
cross_sent = "Yes" if CROSS_SENTENCE else "No"
results_file = './Results/Results.pickle'
if os.path.isfile(results_file):
df = pd.read_pickle(results_file)
else:
df = pd.DataFrame()
df = df.append(pd.DataFrame({
"Embedding": embedding_type,
"Method": method,
"Time": datetime.datetime.now(),
"Dimension": dim,
"Window": ww,
"Word count": corpus_size,
"Sampling": SAMPLING,
"Cross-sentence": cross_sent,
"Epochs": str(EPOCHS),
"Dataset": dataset_name,
"Out of vocabulary": out_of_vocab,
"Low bin lower bound": low_bounds[0],
"Low bin upper bound": low_bounds[1],
"Mid bin lower bound": mid_bounds[0],
"Mid bin upper bound": mid_bounds[1],
"High bin lower bound": high_bounds[0],
"High bin upper bound": high_bounds[1],
"Low bin score": low_similarity,
"Low bin pair count": low_count,
"Middle bin score": mid_similarity,
"Middle bin pair count": mid_count,
"High bin score": high_similarity,
"High bin pair count": high_count,
"Mixed bin score": mixed_similarity,
"Mixed bin pair count": mixed_count,
"General score": gen_similarity,
}, index=[0]), ignore_index=True)
df.to_pickle(results_file)
print(df)
def save_analogy_results(embedding_type, score, dim, ww, dataset_name, corpus_size):
method = "SG" if SKIP_GRAMS else "CBOW"
cross_sent = "Yes" if CROSS_SENTENCE else "No"
results_file = './Results/Results_analogy.pickle'
if os.path.isfile(results_file):
df = pd.read_pickle(results_file)
else:
df = pd.DataFrame()
df = df.append(pd.DataFrame({
"Embedding": embedding_type,
"Method": method,
"Time": datetime.datetime.now(),
"Dimension": dim,
"Window": ww,
"Word count": corpus_size,
"Sampling": SAMPLING,
"Cross-sentence": cross_sent,
"Epochs": str(EPOCHS),
"Dataset": dataset_name,
"Score": score,
}, index=[0]), ignore_index=True)
df.to_pickle(results_file)
print(df)
def evaluate_w2v(data, current_model, similarity_pairs):
general_similarity = evaluate_similarity(current_model, data.X, data.y)
low_similarity = evaluate_similarity(current_model,
np.asarray(similarity_pairs['low'][0]),
np.asarray(similarity_pairs['low'][1])
)
mid_similarity = evaluate_similarity(current_model,
np.asarray(similarity_pairs['mid'][0]),
np.asarray(similarity_pairs['mid'][1])
)
high_similarity = evaluate_similarity(current_model,
np.asarray(similarity_pairs['high'][0]),
np.asarray(similarity_pairs['high'][1])
)
mixed_similarity = evaluate_similarity(current_model,
np.asarray(similarity_pairs['mixed'][0]),
np.asarray(similarity_pairs['mixed'][1])
)
return general_similarity, low_similarity, mid_similarity, high_similarity, mixed_similarity
def evaluate_fasttext(current_model, X, y):
oov_pairs = 0
for query in X:
out_of_vocabulary = False
for query_word in query:
if query_word not in current_model.wv.vocab:
out_of_vocabulary = True
if out_of_vocabulary:
oov_pairs += 1;
A = np.vstack(current_model.wv[word] for word in X[:, 0])
B = np.vstack(current_model.wv[word] for word in X[:, 1])
scores = np.array([v1.dot(v2.T) / (np.linalg.norm(v1) * np.linalg.norm(v2)) for v1, v2 in zip(A, B)])
return scipy.stats.spearmanr(scores, y).correlation, oov_pairs
def evaluate_w2v_analogy(datset, data, sentences, vocabulary, emb_model, model_dimension, model_window, model_wordcount):
model_results = evaluate_analogy(embedding_model, data.X, data.y)
save_analogy_results('w2v', model_results, model_dimension, model_window, datset, model_wordcount)
def evaluation(datset, data, sentences, vocabulary, emb_model, model_type, model_dimension, model_window, model_wordcount):
high, mid, low, high_bounds, mid_bounds, low_bounds, out_of_voc = single_word_bins(sentences, vocabulary)
pair_similarity = pair_bins(low, mid, high, data)
if model_type == 'w2v':
model_general_similarity, model_low_similarity, model_mid_similarity, \
model_high_similarity, model_mixed_similarity = evaluate_w2v(data, emb_model, pair_similarity)
save_results('w2v', model_general_similarity, model_low_similarity, model_mid_similarity, model_high_similarity,
model_mixed_similarity, len(out_of_voc), len(pair_similarity['low'][0]),
len(pair_similarity['mid'][0]), len(pair_similarity['high'][0]),
len(pair_similarity['mixed'][0]), low_bounds, mid_bounds, high_bounds, model_dimension,
model_window, datset, model_wordcount)
else:
model_general_similarity, out_of_voc = evaluate_fasttext(emb_model, data.X, data.y)
model_low_similarity, buffer = evaluate_fasttext(emb_model,
np.asarray(pair_similarity['low'][0]),
np.asarray(pair_similarity['low'][1])
)
model_mid_similarity, buffer = evaluate_fasttext(emb_model,
np.asarray(pair_similarity['mid'][0]),
np.asarray(pair_similarity['mid'][1])
)
model_high_similarity, buffer = evaluate_fasttext(emb_model,
np.asarray(pair_similarity['high'][0]),
np.asarray(pair_similarity['high'][1])
)
model_mixed_similarity, buffer = evaluate_fasttext(emb_model,
np.asarray(pair_similarity['mixed'][0]),
np.asarray(pair_similarity['mixed'][1])
)
save_results('fasttext', model_general_similarity, model_low_similarity, model_mid_similarity, model_high_similarity,
model_mixed_similarity, out_of_voc, len(pair_similarity['low'][0]),
len(pair_similarity['mid'][0]), len(pair_similarity['high'][0]),
len(pair_similarity['mixed'][0]), low_bounds, mid_bounds, high_bounds, model_dimension,
model_window, datset, model_wordcount)
corpus = api.load("wiki-english-20171001")
for word_count in SIZES:
sents = corpus_processing(corpus, word_count, RANDOMIZE_ARTICLES)
for embedding in USE_EMBEDDING:
for dimension in DIMENSION:
for word_window in WORD_WINDOWS:
model_filename = "./Models/" + embedding + "_" + SG_MAP[SKIP_GRAMS] + "_" \
+ SIZES_MAP[word_count] + "_" + str(dimension) + "_" + str(word_window) + "_" \
+ SAMPLING + "_" + str(EPOCHS)
vocab_filename = "./Vocab/" + embedding + "_" + SG_MAP[SKIP_GRAMS] + "_" \
+ SIZES_MAP[word_count] + "_" + str(dimension) + "_" + str(word_window) + "_" \
+ SAMPLING + "_" + str(EPOCHS)
if not USE_CACHED:
skip_grams = 1 if SKIP_GRAMS else 0
softmax = 1 if SAMPLING == 'hs' else 0
negative_sample = 10 if SAMPLING == 'ns' else 0
if embedding == 'w2v':
model = Word2Vec(sents, size = dimension, window = word_window, workers=3,
sg = skip_grams, hs = softmax, negative = negative_sample, iter = EPOCHS)
model.wv.save_word2vec_format(model_filename, binary=False)
model.vocabulary.save(vocab_filename)
else:
model = FastText(sg = skip_grams, hs = softmax, size = dimension, window = word_window,
workers = 3, negative = negative_sample)
model.build_vocab(sentences = sents)
model.train(sentences = sents, total_examples = len(sents), epochs = EPOCHS)
model.save(model_filename)
# model.vocabulary.save(vocab_filename)
print("Model saved at " + model_filename)
embedding_model = []
if embedding == 'w2v':
embedding_model = Embedding.from_word2vec(model_filename)
else:
embedding_model = FastText.load(model_filename)
for dset in USE_DATASETS:
dataset = get_dataset(dset)
vocab = set()
for pair in dataset.X:
vocab.add(pair[0])
vocab.add(pair[1])
if dset in ['Google', 'MSR']:
evaluate_w2v_analogy(dset, dataset, sents, vocab, embedding_model, dimension, word_window, word_count)
else:
evaluation(dset, dataset, sents, vocab, embedding_model, embedding, dimension, word_window, word_count)
# high, mid, low, high_bounds, mid_bounds, low_bounds, out_of_voc = single_word_bins(sents, vocab)
# pair_similarity = pair_bins(low, mid, high, dataset)
#
# model_general_similarity = evaluate_similarity(model, dataset.X, dataset.y)
# model_low_similarity = evaluate_similarity(model,
# np.asarray(pair_similarity['low'][0]),
# np.asarray(pair_similarity['low'][1])
# )
# model_mid_similarity = evaluate_similarity(model,
# np.asarray(pair_similarity['mid'][0]),
# np.asarray(pair_similarity['mid'][1])
# )
# model_high_similarity = evaluate_similarity(model,
# np.asarray(pair_similarity['high'][0]),
# np.asarray(pair_similarity['high'][1])
# )
# model_mixed_similarity = evaluate_similarity(model,
# np.asarray(pair_similarity['mixed'][0]),
# np.asarray(pair_similarity['mixed'][1])
# )
# save_results(embedding, model_general_similarity, model_low_similarity, model_mid_similarity,
# model_high_similarity,
# model_mixed_similarity, len(out_of_voc), len(pair_similarity['low'][0]),
# len(pair_similarity['mid'][0]), len(pair_similarity['high'][0]),
# len(pair_similarity['mixed'][0]), low_bounds, mid_bounds, high_bounds, dimension, word_window,
# dset, word_count)
# if embedding == 'w2v':
# evaluate_w2v(dset, dataset, sents, vocab, embedding_model, dimension, word_window, word_count)
# else
# evaluate_fasttext(dset, dataset, sents, vocab, embedding_model, dimension, word_window, word_count)