forked from syxu828/Crosslingula-KG-Matching
-
Notifications
You must be signed in to change notification settings - Fork 0
/
aggregators.py
439 lines (326 loc) · 14.8 KB
/
aggregators.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
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
import tensorflow as tf
from layers import Layer, Dense
from inits import glorot, zeros, random
from pooling import mean_pool
from match_utils import multi_highway_layer
class GatedMeanAggregator(Layer):
def __init__(self, input_dim, output_dim, neigh_input_dim=None,
dropout=0, bias=True, act=tf.nn.relu,
name=None, concat=False, **kwargs):
super(GatedMeanAggregator, self).__init__(**kwargs)
self.dropout = dropout
self.bias = bias
self.act = act
self.concat = concat
if name is not None:
name = '/' + name
else:
name = ''
if neigh_input_dim == None:
neigh_input_dim = input_dim
if concat:
self.output_dim = 2 * output_dim
with tf.variable_scope(self.name + name + '_vars'):
self.vars['neigh_weights'] = glorot([neigh_input_dim, output_dim],
name='neigh_weights')
self.vars['self_weights'] = glorot([input_dim, output_dim],
name='self_weights')
if self.bias:
self.vars['bias'] = zeros([self.output_dim], name='bias')
self.vars['gate_weights'] = glorot([2*output_dim, 2*output_dim],
name='gate_weights')
self.vars['gate_bias'] = zeros([2*output_dim], name='bias')
self.input_dim = input_dim
self.output_dim = output_dim
def _call(self, inputs):
self_vecs, neigh_vecs = inputs
neigh_vecs = tf.nn.dropout(neigh_vecs, 1-self.dropout)
self_vecs = tf.nn.dropout(self_vecs, 1-self.dropout)
neigh_means = tf.reduce_mean(neigh_vecs, axis=1)
# [nodes] x [out_dim]
from_neighs = tf.matmul(neigh_means, self.vars['neigh_weights'])
from_self = tf.matmul(self_vecs, self.vars["self_weights"])
if not self.concat:
output = tf.add_n([from_self, from_neighs])
else:
output = tf.concat([from_self, from_neighs], axis=1)
# bias
if self.bias:
output += self.vars['bias']
gate = tf.concat([from_self, from_neighs], axis=1)
gate = tf.matmul(gate, self.vars["gate_weights"]) + self.vars["gate_bias"]
gate = tf.nn.relu(gate)
return gate*self.act(output)
class GCNAggregator(Layer):
"""
Aggregates via mean followed by matmul and non-linearity.
Same matmul parameters are used self vector and neighbor vectors.
"""
def __init__(self, input_dim, output_dim, neigh_input_dim=None,
dropout=0., bias=True, act=tf.nn.relu, name=None, concat=False, mode="train", **kwargs):
super(GCNAggregator, self).__init__(**kwargs)
self.dropout = dropout
self.bias = bias
self.act = act
self.concat = concat
self.mode = mode
if neigh_input_dim is None:
neigh_input_dim = input_dim
if name is not None:
name = '/' + name
else:
name = ''
with tf.variable_scope(self.name + name + '_vars'):
self.vars['weights'] = glorot([neigh_input_dim, output_dim],
name='neigh_weights')
if self.bias:
self.vars['bias'] = zeros([output_dim], name='bias')
if self.logging:
self._log_vars()
self.input_dim = input_dim
self.output_dim = output_dim
def _call(self, inputs):
self_vecs, neigh_vecs = inputs
if self.mode == "train":
neigh_vecs = tf.nn.dropout(neigh_vecs, 1 - self.dropout)
self_vecs = tf.nn.dropout(self_vecs, 1 - self.dropout)
means = tf.reduce_mean(tf.concat([neigh_vecs, tf.expand_dims(self_vecs, axis=1)], axis=1), axis=1)
# [nodes] x [out_dim]
output = tf.matmul(means, self.vars['weights'])
# bias
if self.bias:
output += self.vars['bias']
return self.act(output), self.output_dim
class MeanAggregator(Layer):
"""Aggregates via mean followed by matmul and non-linearity."""
def __init__(self, input_dim, output_dim, neigh_input_dim=None,
dropout=0, bias=True, act=tf.nn.relu,
name=None, concat=False, mode="train", if_use_high_way=False, **kwargs):
super(MeanAggregator, self).__init__(**kwargs)
self.dropout = dropout
self.bias = bias
self.act = act
self.concat = concat
self.mode = mode
self.if_use_high_way = if_use_high_way
if name is not None:
name = '/' + name
else:
name = ''
if neigh_input_dim == None:
neigh_input_dim = input_dim
self.neigh_input_dim = neigh_input_dim
if concat:
self.output_dim = 2 * output_dim
else:
self.output_dim = output_dim
with tf.variable_scope(self.name + name + '_vars'):
self.vars['neigh_weights'] = glorot([neigh_input_dim, output_dim],
name='neigh_weights')
self.vars['self_weights'] = glorot([input_dim, output_dim],
name='self_weights')
# self.vars['neigh_weights'] = random([neigh_input_dim, output_dim], name='neigh_weights')
# self.vars['self_weights'] = random([input_dim, output_dim], name='neigh_weights')
if self.bias:
self.vars['bias'] = zeros([self.output_dim], name='bias')
self.input_dim = input_dim
self.output_dim = output_dim
if self.concat:
self.output_dim = output_dim * 2
def _call(self, inputs):
self_vecs, neigh_vecs = inputs
if self.mode == "train":
neigh_vecs = tf.nn.dropout(neigh_vecs, 1-self.dropout)
self_vecs = tf.nn.dropout(self_vecs, 1-self.dropout)
# reduce_mean performs better than mean_pool
neigh_means = tf.reduce_mean(neigh_vecs, axis=1)
# neigh_means = mean_pool(neigh_vecs, neigh_len)
# [nodes] x [out_dim]
from_neighs = tf.matmul(neigh_means, self.vars['neigh_weights'])
if self.if_use_high_way:
with tf.variable_scope("fw_hidden_highway"):
fw_hidden = multi_highway_layer(from_neighs, self.neigh_input_dim, 1)
from_self = tf.matmul(self_vecs, self.vars["self_weights"])
if not self.concat:
output = tf.add_n([from_self, from_neighs])
else:
output = tf.concat([from_self, from_neighs], axis=1)
# bias
if self.bias:
output += self.vars['bias']
return self.act(output), self.output_dim
class AttentionAggregator(Layer):
""" Attention-based aggregator """
def __init__(self, input_dim, output_dim, neigh_input_dim=None,
dropout=0, bias=True, act=tf.nn.relu,
name=None, concat=False, mode="train", **kwargs):
super(AttentionAggregator, self).__init__(**kwargs)
self.dropout = dropout
self.bias = bias
self.act = act
self.concat = concat
self.mode = mode
if name is not None:
name = '/' + name
else:
name = ''
if neigh_input_dim == None:
neigh_input_dim = input_dim
self.input_dim = input_dim
self.output_dim = output_dim
with tf.variable_scope(self.name + name + '_vars'):
if self.bias:
self.vars['bias'] = zeros([self.output_dim], name='bias')
self.q_dense_layer = Dense(input_dim=input_dim, output_dim=input_dim, bias=False, sparse_inputs=False,
name="q")
self.k_dense_layer = Dense(input_dim=input_dim, output_dim=input_dim, bias=False, sparse_inputs=False,
name="k")
self.v_dense_layer = Dense(input_dim=input_dim, output_dim=input_dim, bias=False, sparse_inputs=False,
name="v")
self.output_dense_layer = Dense(input_dim=input_dim, output_dim=output_dim, bias=False, sparse_inputs=False, name="output_transform")
def _call(self, inputs):
self_vecs, neigh_vecs= inputs
q = self.q_dense_layer(self_vecs)
neigh_vecs = tf.concat([tf.expand_dims(self_vecs, axis=1), neigh_vecs], axis=1)
neigh_len = tf.shape(neigh_vecs)[1]
neigh_vecs = tf.reshape(neigh_vecs, [-1, self.input_dim])
k = self.k_dense_layer(neigh_vecs)
v = self.v_dense_layer(neigh_vecs)
k = tf.reshape(k, [-1, neigh_len, self.input_dim])
v = tf.reshape(v, [-1, neigh_len, self.input_dim])
logits = tf.reduce_sum(tf.multiply(tf.expand_dims(q, axis=1), k), axis=-1)
# if self.bias:
# logits += self.vars['bias']
weights = tf.nn.softmax(logits, name="attention_weights")
attention_output = tf.reduce_sum(tf.multiply(tf.expand_dims(weights, axis=-1), v), axis=1)
attention_output = self.output_dense_layer(attention_output)
return attention_output, self.output_dim
class MaxPoolingAggregator(Layer):
""" Aggregates via max-pooling over MLP functions."""
def __init__(self, input_dim, output_dim, model_size="small", neigh_input_dim=None,
dropout=0., bias=True, act=tf.nn.relu, name=None, concat=False, mode="train", **kwargs):
super(MaxPoolingAggregator, self).__init__(**kwargs)
self.mode = mode
self.dropout = dropout
self.bias = bias
self.act = act
self.concat = concat
if name is not None:
name = '/' + name
else:
name = ''
if neigh_input_dim == None:
neigh_input_dim = input_dim
if concat:
self.output_dim = 2 * output_dim
if model_size == "small":
hidden_dim = self.hidden_dim = 50
elif model_size == "big":
hidden_dim = self.hidden_dim = 50
self.mlp_layers = []
self.mlp_layers.append(Dense(input_dim=neigh_input_dim, output_dim=hidden_dim, act=tf.nn.relu,
dropout=dropout, sparse_inputs=False, logging=self.logging))
with tf.variable_scope(self.name + name + '_vars'):
self.vars['neigh_weights'] = glorot([hidden_dim, output_dim], name='neigh_weights')
self.vars['self_weights'] = glorot([input_dim, output_dim], name='self_weights')
if self.bias:
self.vars['bias'] = zeros([self.output_dim], name='bias')
self.input_dim = input_dim
self.output_dim = output_dim
self.neigh_input_dim = neigh_input_dim
if self.concat:
self.output_dim = output_dim * 2
def _call(self, inputs):
self_vecs, neigh_vecs = inputs
neigh_h = neigh_vecs
dims = tf.shape(neigh_h)
batch_size = dims[0]
num_neighbors = dims[1]
h_reshaped = tf.reshape(neigh_h, (batch_size * num_neighbors, self.neigh_input_dim))
for l in self.mlp_layers:
h_reshaped = l(h_reshaped)
neigh_h = tf.reshape(h_reshaped, (batch_size, num_neighbors, self.hidden_dim))
neigh_h = tf.reduce_max(neigh_h, axis=1)
from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights'])
from_self = tf.matmul(self_vecs, self.vars["self_weights"])
if not self.concat:
output = tf.add_n([from_self, from_neighs])
else:
output = tf.concat([from_self, from_neighs], axis=1)
# bias
if self.bias:
output += self.vars['bias']
return self.act(output), self.output_dim
class SeqAggregator(Layer):
""" Aggregates via a standard LSTM.
"""
def __init__(self, input_dim, output_dim, model_size="small", neigh_input_dim=None,
dropout=0., bias=True, act=tf.nn.relu, name=None, concat=False, mode="train", **kwargs):
super(SeqAggregator, self).__init__(**kwargs)
self.dropout = dropout
self.bias = bias
self.act = act
self.concat = concat
self.mode = mode
self.output_dim = output_dim
if neigh_input_dim is None:
neigh_input_dim = input_dim
if name is not None:
name = '/' + name
else:
name = ''
if model_size == "small":
hidden_dim = self.hidden_dim = 128
elif model_size == "big":
hidden_dim = self.hidden_dim = 256
with tf.variable_scope(self.name + name + '_vars'):
self.vars['neigh_weights'] = glorot([hidden_dim, output_dim],
name='neigh_weights')
self.vars['self_weights'] = glorot([input_dim, output_dim],
name='self_weights')
if self.bias:
self.vars['bias'] = zeros([self.output_dim], name='bias')
if self.logging:
self._log_vars()
self.input_dim = input_dim
self.output_dim = output_dim
self.neigh_input_dim = neigh_input_dim
self.cell = tf.contrib.rnn.BasicLSTMCell(self.hidden_dim)
def _call(self, inputs):
self_vecs, neigh_vecs = inputs
dims = tf.shape(neigh_vecs)
batch_size = dims[0]
initial_state = self.cell.zero_state(batch_size, tf.float32)
used = tf.sign(tf.reduce_max(tf.abs(neigh_vecs), axis=2))
length = tf.reduce_sum(used, axis=1)
length = tf.maximum(length, tf.constant(1.))
length = tf.cast(length, tf.int32)
with tf.variable_scope(self.name) as scope:
try:
rnn_outputs, rnn_states = tf.nn.dynamic_rnn(
self.cell, neigh_vecs,
initial_state=initial_state, dtype=tf.float32, time_major=False,
sequence_length=length)
except ValueError:
scope.reuse_variables()
rnn_outputs, rnn_states = tf.nn.dynamic_rnn(
self.cell, neigh_vecs,
initial_state=initial_state, dtype=tf.float32, time_major=False,
sequence_length=length)
batch_size = tf.shape(rnn_outputs)[0]
max_len = tf.shape(rnn_outputs)[1]
out_size = int(rnn_outputs.get_shape()[2])
index = tf.range(0, batch_size) * max_len + (length - 1)
flat = tf.reshape(rnn_outputs, [-1, out_size])
neigh_h = tf.gather(flat, index)
from_neighs = tf.matmul(neigh_h, self.vars['neigh_weights'])
from_self = tf.matmul(self_vecs, self.vars["self_weights"])
output = tf.add_n([from_self, from_neighs])
if not self.concat:
output = tf.add_n([from_self, from_neighs])
else:
output = tf.concat([from_self, from_neighs], axis=1)
# bias
if self.bias:
output += self.vars['bias']
return self.act(output)