-
Notifications
You must be signed in to change notification settings - Fork 4
/
comp_models.py
401 lines (334 loc) · 16.2 KB
/
comp_models.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
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
from model.models import *
import tensorflow as tf
import model
from tensorflow.models.rnn.rnn_cell import *
class CompositionalKBScoringModel(AbstractKBScoringModel):
def __init__(self, kb, size, batch_size, comp_model, is_train=True, num_neg=200, learning_rate=1e-2):
self._comp_model = comp_model
AbstractKBScoringModel.__init__(self, kb, size, batch_size, is_train, num_neg, learning_rate, 0.0, False)
def _input_params(self):
return [self._rel_input]
def _init_inputs(self):
self._rel_input = tf.placeholder(tf.float32, shape=[None, self._size], name="rel")
self._subj_input = tf.placeholder(tf.int64, shape=[None], name="subj")
self._obj_input = tf.placeholder(tf.int64, shape=[None], name="obj")
self._subj_in = np.zeros([self._batch_size], dtype=np.int64)
self._obj_in = np.zeros([self._batch_size], dtype=np.int64)
self._rel_in = np.zeros([self._batch_size, self._size], dtype=np.float32)
self._feed_dict = {}
def _start_adding_triples(self):
self._rels = []
def _add_triple_to_input(self, t, j):
(rel, subj, obj) = t
self._subj_in[j] = self._kb.get_id(subj, 1)
self._obj_in[j] = self._kb.get_id(obj, 2)
self._rels.append(rel)
def _finish_adding_triples(self, batch_size):
if batch_size < self._batch_size:
self._feed_dict[self._subj_input] = self._subj_in[:batch_size]
self._feed_dict[self._obj_input] = self._obj_in[:batch_size]
self._feed_dict[self._rel_input] = self._rel_in[:batch_size]
else:
self._feed_dict[self._subj_input] = self._subj_in
self._feed_dict[self._obj_input] = self._obj_in
self._feed_dict[self._rel_input] = self._rel_in
def name(self):
return self.__class__.__name__ + "__" + self._comp_model.name()
def _get_feed_dict(self):
return self._feed_dict
def _composition_forward(self, sess):
rel_embeddings = self._comp_model.forward(sess, self._rels)
for b in xrange(len(rel_embeddings)):
self._rel_in[b] = rel_embeddings[b]
def _composition_backward(self, sess, grads):
grad_list = [grads[0][b] for b in xrange(grads[0].shape[0])]
self._comp_model.backward(sess, grad_list)
def score_triples(self, sess, triples):
i = 0
result = np.zeros([len(triples)])
while i < len(triples):
batch_size = min(self._batch_size, len(triples)-i)
self._start_adding_triples()
for j in xrange(batch_size):
self._add_triple_to_input(triples[i+j], j)
self._finish_adding_triples(batch_size)
self._composition_forward(sess)
result[i:i+batch_size] = sess.run(self._scores, feed_dict=self._get_feed_dict())
i += batch_size
return result
def step(self, sess, pos_triples, neg_triples, mode="update"):
'''
:param sess: tf session
:param pos_triples: list of positive triple
:param neg_triples: list of (lists of) negative triples
:param mode: default(train)|loss|accumulate(used for batch training)
:return:
'''
assert self._is_train, "model has to be created in training mode!"
assert len(pos_triples) + reduce(lambda acc, x: acc+len(x), neg_triples, 0) == self._batch_size, \
"batch_size and provided batch do not fit"
j = 0
self._start_adding_triples()
for pos, negs in zip(pos_triples, neg_triples):
self._add_triple_to_input(pos, j)
j += 1
for neg in negs:
self._add_triple_to_input(neg, j)
j += 1
self._finish_adding_triples(j)
self._composition_forward(sess)
if mode == "loss":
return sess.run(self._loss, feed_dict=self._get_feed_dict())
else:
assert self._is_train, "training only possible in training state."
if hasattr(self, "_update"):
res = sess.run([self._loss, self._update] + self._input_grads, feed_dict=self._get_feed_dict())
self._composition_backward(sess, res[2:])
else:
res = sess.run([self._loss] + self._input_grads, feed_dict=self._get_feed_dict())
self._composition_backward(sess, res[1:])
return res[0]
class CompDistMult(CompositionalKBScoringModel):
def _scoring_f(self):
with tf.device("/cpu:0"):
E_subjs = tf.get_variable("E_s", [len(self._kb.get_symbols(1)), self._size])
E_objs = tf.get_variable("E_o", [len(self._kb.get_symbols(2)), self._size])
self.e_subj = tf.tanh(tf.nn.embedding_lookup(E_subjs, self._subj_input))
self.e_obj = tf.tanh(tf.nn.embedding_lookup(E_objs, self._obj_input))
self.e_rel = self._rel_input # relation is already embedded by composition function
s_o_prod = self.e_obj * self.e_subj
score = tf_util.batch_dot(self.e_rel, s_o_prod)
return score
class CompModelE(CompositionalKBScoringModel):
def _init_inputs(self):
self._rel_input = tf.placeholder(tf.float32, shape=[None, 2*self._size], name="rel")
self._subj_input = tf.placeholder(tf.int64, shape=[None], name="subj")
self._obj_input = tf.placeholder(tf.int64, shape=[None], name="obj")
self._subj_in = np.zeros([self._batch_size], dtype=np.int64)
self._obj_in = np.zeros([self._batch_size], dtype=np.int64)
self._rel_in = np.zeros([self._batch_size, 2*self._size], dtype=np.float32)
self._feed_dict = {}
def _scoring_f(self):
with tf.device("/cpu:0"):
E_subjs = tf.get_variable("E_s", [len(self._kb.get_symbols(1)), self._size])
E_objs = tf.get_variable("E_o", [len(self._kb.get_symbols(2)), self._size])
self.e_rel_s, self.e_rel_o = tf.split(1, 2, self._rel_input)
self.e_subj = tf.tanh(tf.nn.embedding_lookup(E_subjs, self._subj_input))
self.e_obj = tf.tanh(tf.nn.embedding_lookup(E_objs, self._obj_input))
score = tf_util.batch_dot(self.e_rel_s, self.e_subj) + tf_util.batch_dot(self.e_rel_o, self.e_obj)
return score
class CompModelO(CompositionalKBScoringModel):
def __init__(self, kb, size, batch_size, comp_model, is_train=True, num_neg=200, learning_rate=1e-2,
which_sets=["train_text"]):
self._which_sets = set(which_sets)
CompositionalKBScoringModel.__init__(self, kb, size, batch_size, comp_model, is_train=True, num_neg=200,
learning_rate=1e-2)
def _init_inputs(self):
self._rel_ids = dict()
len(self._kb.get_symbols(0))
# create tuple to rel lookup
self._tuple_rels_lookup = dict()
for (rel, subj, obj), _, typ in self._kb.get_all_facts():
if typ in self._which_sets:
s_i = self._kb.get_id(subj, 1)
o_i = self._kb.get_id(obj, 2)
t = (s_i, o_i)
if t not in self._tuple_rels_lookup:
self._tuple_rels_lookup[t] = [rel]
else:
self._tuple_rels_lookup[t].append(rel)
t = (o_i, s_i)
if t not in self._tuple_rels_lookup:
self._tuple_rels_lookup[t] = [rel+"_inv"]
else:
self._tuple_rels_lookup[t].append(rel+"_inv")
self._rel_input = tf.placeholder(tf.float32, shape=[None, self._size], name="rel")
self._rel_in = np.zeros([self._batch_size, self._size], dtype=np.float32)
self._observed_input = tf.placeholder(tf.float32, shape=[None, self._size], name="observed")
self._observed_in = np.zeros([self._batch_size, self._size], dtype=np.float32)
self._feed_dict = {}
def _start_adding_triples(self):
self._rels = []
self.__offsets = []
def _add_triple_to_input(self, t, j):
self.__offsets.append(len(self._rels))
(rel, subj, obj) = t
self._rels.append(rel)
s_i = self._kb.get_id(subj, 1)
o_i = self._kb.get_id(obj, 2)
rels = self._tuple_rels_lookup.get((s_i, o_i))
if rels:
for i in xrange(len(rels)):
if rels[i] != rel:
self._rels.append(rels[i])
def _finish_adding_triples(self, batch_size):
if batch_size < self._batch_size:
self._feed_dict[self._rel_input] = self._rel_in[:batch_size]
self._feed_dict[self._observed_input] = self._observed_in[:batch_size]
else:
self._feed_dict[self._rel_input] = self._rel_in
self._feed_dict[self._observed_input] = self._observed_in
def _scoring_f(self):
return tf_util.batch_dot(self._rel_input, self._observed_input)
def _input_params(self):
return [self._rel_input, self._observed_input]
def _composition_forward(self, sess):
rel_embeddings = self._comp_model.forward(sess, self._rels)
for b, off in enumerate(self.__offsets):
self._rel_in[b] = rel_embeddings[off]
end = self.__offsets[b+1] if len(self.__offsets) > (b+1) else len(self._rels)
self._observed_in[b] *= 0.0
for i in xrange(off+1, end):
self._observed_in[b] += rel_embeddings[i]
if (end-off-1) > 0:
self._observed_in[b] /= (end-off-1)
def _composition_backward(self, sess, grads):
grad_list = []
for b, off in enumerate(self.__offsets):
grad_list.append(grads[0][b])
end = self.__offsets[b+1] if len(self.__offsets) > (b+1) else len(self._rels)
observed_grad = grads[1][b]
if (end-off-1) > 0:
observed_grad /= (end-off-1)
for i in xrange(off+1, end):
grad_list.append(observed_grad)
self._comp_model.backward(sess, grad_list)
class CompWeightedModelO(CompModelO):
def _init_inputs(self):
self._rel_ids = dict()
len(self._kb.get_symbols(0))
# create tuple to rel lookup
self._tuple_rels_lookup = dict()
for (rel, subj, obj), _, typ in self._kb.get_all_facts():
if typ in self._which_sets:
s_i = self._kb.get_id(subj, 1)
o_i = self._kb.get_id(obj, 2)
t = (s_i, o_i)
if t not in self._tuple_rels_lookup:
self._tuple_rels_lookup[t] = [rel]
else:
self._tuple_rels_lookup[t].append(rel)
t = (o_i, s_i)
if t not in self._tuple_rels_lookup:
self._tuple_rels_lookup[t] = [rel+"_inv"]
else:
self._tuple_rels_lookup[t].append(rel+"_inv")
self._rel_input = tf.placeholder(tf.float32, shape=[None, self._size], name="rel")
self._sparse_indices_input = tf.placeholder(tf.int64, name="sparse_indices")
self._shape_input = tf.placeholder(tf.int64, name="shape")
self._observed_input = tf.placeholder(tf.float32, [None, self._size], name="observation_inputs")
self._feed_dict = {}
def _start_adding_triples(self):
self._sparse_indices = []
self._rel_in = []
self._rels = []
self._obs_in = []
self.__offsets = []
self._max_cols = 1
def _add_triple_to_input(self, t, b):
self.__offsets.append(len(self._rels))
(rel, subj, obj) = t
s_i = self._kb.get_id(subj, 1)
o_i = self._kb.get_id(obj, 2)
rels = self._tuple_rels_lookup.get((s_i, o_i))
if rels and any(rel_i != rel for rel_i in rels):
self._rels.append(rel)
for i in xrange(len(rels)):
if rels[i] != rel:
self._sparse_indices.append([b, i])
self._rels.append(rels[i])
self._max_cols = max(self._max_cols, len(rels))
else:
self._sparse_indices.append([b, 0])
def _finish_adding_triples(self, batch_size):
self._feed_dict[self._sparse_indices_input] = self._sparse_indices
self._feed_dict[self._shape_input] = [batch_size, self._max_cols]
self._feed_dict[self._observed_input] = self._obs_in
self._feed_dict[self._rel_input] = self._rel_in
def _scoring_f(self):
scores_flat = tf_util.batch_dot(self._rel_input, self._observed_input)
# for softmax set empty cells to something very small, so weight becomes practically zero
scores = tf.sparse_to_dense(self._sparse_indices_input, self._shape_input,
scores_flat, default_value=-1e-3)
softmax = tf.nn.softmax(scores)
weighted_scores = tf.reduce_sum(scores * softmax, reduction_indices=[1], keep_dims=False)
return weighted_scores
def _composition_forward(self, sess):
rel_embeddings = self._comp_model.forward(sess, self._rels)
zero_v = np.zeros([self._size], dtype=np.float32)
for b, off in enumerate(self.__offsets):
end = self.__offsets[b+1] if len(self.__offsets) > (b+1) else len(self._rels)
if end == off:
self._rel_in.append(zero_v) # default relation if none was observed
self._obs_in.append(zero_v)
else:
for i in xrange(end - (off+1)):
self._rel_in.append(rel_embeddings[off])
self._obs_in.extend(rel_embeddings[off+1:end])
def _composition_backward(self, sess, grads):
grad_list = []
skip = 0
for b, off in enumerate(self.__offsets):
end = self.__offsets[b+1] if len(self.__offsets) > (b+1) else len(self._rels)
if end > off:
off = off - b + skip
end = end - b + skip
rel_grad = grads[0][off]
for i in xrange(off+1, end):
rel_grad += grads[0][i]
grad_list.append(rel_grad)
observed_grads = grads[1][off:end]
grad_list.extend(observed_grads)
else:
skip += 1
self._comp_model.backward(sess, grad_list)
class CompCombinedModel(CompositionalKBScoringModel):
def __init__(self, models, kb, size, batch_size, is_train=True, num_neg=200, learning_rate=1e-2, l2_lambda=0.0,
is_batch_training=False, composition=None, share_vars=False):
self._models = []
self.__name = '_'.join(models)
if composition:
self.__name = composition + "__" + self.__name
with vs.variable_scope(self.name()):
for m in models:
self._models.append(model.create_model(kb, size, batch_size, False, num_neg, learning_rate,
l2_lambda, False, composition=composition, type=m))
AbstractKBScoringModel.__init__(self, kb, size, batch_size, is_train, num_neg, learning_rate,
l2_lambda, is_batch_training)
def name(self):
return self.__name
def _scoring_f(self):
weights = map(lambda _: tf.Variable(float(1)), xrange(len(self._models)-1))
scores = [self._models[0]._scores]
for i in xrange(len(self._models)-1):
scores.append(self._models[i+1]._scores * weights[i])
return tf.reduce_sum(tf.pack(scores), 0)
def _add_triple_to_input(self, t, j):
for m in self._models:
m._add_triple_to_input(t, j)
def _finish_adding_triples(self, batch_size):
self._rels = []
for m in self._models:
m._finish_adding_triples(batch_size)
self._feed_dict.update(m._get_feed_dict())
if m._rels:
self._rels.extend(m._rels)
def _start_adding_triples(self):
for m in self._models:
m._start_adding_triples()
def _input_params(self):
ips = []
for m in self._models:
ips.extend(m._input_params())
return ips
def _composition_forward(self, sess):
for m in self._models:
m._composition_forward(sess)
def _composition_backward(self, sess, grads):
i = 0
for m in self._models:
inp_params = m._input_params()
if inp_params:
j = len(inp_params)
m._composition_backward(sess, grads[i:i+j])
i += j