def get_batch_generator(self):
if self.is_train:
while True:
X1, X1_len, X2, X2_len, Y, ID_pairs = self.get_batch()
if self.config['use_dpool']:
yield ({
'query':
X1,
'query_len':
X1_len,
'doc':
X2,
'doc_len':
X2_len,
'dpool_index':
DynamicMaxPooling.dynamic_pooling_index(
X1_len, X2_len, self.config['text1_maxlen'],
self.config['text2_maxlen'])
}, Y)
else:
yield ({
'query': X1,
'query_len': X1_len,
'doc': X2,
'doc_len': X2_len
}, Y)
else:
while self.point + self.batch_size <= self.total_rel_num:
X1, X1_len, X2, X2_len, Y, ID_pairs = self.get_batch(
randomly=False)
if self.config['use_dpool']:
yield ({
'query':
X1,
'query_len':
X1_len,
'doc':
X2,
'doc_len':
X2_len,
'dpool_index':
DynamicMaxPooling.dynamic_pooling_index(
X1_len, X2_len, self.config['text1_maxlen'],
self.config['text2_maxlen']),
'ID':
ID_pairs
}, Y)
else:
yield ({
'query': X1,
'query_len': X1_len,
'doc': X2,
'doc_len': X2_len,
'ID': ID_pairs
}, Y)
def get_batch_generator(self):
for X1, X1_len, X2, X2_len, Y, ID_pairs, list_counts in self.get_batch(
):
if self.config['use_dpool']:
yield ({
'query':
X1,
'query_len':
X1_len,
'doc':
X2,
'doc_len':
X2_len,
'dpool_index':
DynamicMaxPooling.dynamic_pooling_index(
X1_len, X2_len, self.config['text1_maxlen'],
self.config['text2_maxlen']),
'ID':
ID_pairs,
'list_counts':
list_counts
}, Y)
else:
yield ({
'query': X1,
'query_len': X1_len,
'doc': X2,
'doc_len': X2_len,
'ID': ID_pairs,
'list_counts': list_counts
}, Y)
def get_batch_generator(self):
while True:
X1, X1_len, X2, X2_len, Y = self.get_batch()
if self.config['use_dpool']:
yield ({
'query':
X1,
'query_len':
X1_len,
'doc':
X2,
'doc_len':
X2_len,
'dpool_index':
DynamicMaxPooling.dynamic_pooling_index(
X1_len, X2_len, self.config['text1_maxlen'],
self.config['text2_maxlen'])
}, Y)
else:
yield ({
'query': X1,
'query_len': X1_len,
'doc': X2,
'doc_len': X2_len
}, Y)
def get_batch_generator(self):
while True:
sample = self.get_batch()
if not sample:
break
X1, X1_len, X2, X2_len, Y, ID_pairs = sample
if self.config['use_dpool']:
yield ({
'query':
X1,
'query_len':
X1_len,
'doc':
X2,
'doc_len':
X2_len,
'dpool_index':
DynamicMaxPooling.dynamic_pooling_index(
X1_len, X2_len, self.config['text1_maxlen'],
self.config['text2_maxlen']),
'ID':
ID_pairs
}, Y)
else:
yield ({
'query': X1,
'query_len': X1_len,
'doc': X2,
'doc_len': X2_len,
'ID': ID_pairs
}, Y)
def get_dpool_index(self, _len1, _len2):
'''
get dynamic pooling index
@param _len1: int length of text1 terms
@param _len2: int length of text2 terms
@return: np.array(index)
'''
_dpool_index_arr = DynamicMaxPooling.dynamic_pooling_index([_len1,], [_len2,], self.config['text1_maxlen'], self.config['text2_maxlen'])
return _dpool_index_arr
def get_batch_generator(self):
while True:
X1, XP1, X1_len, XP1_len, X2, XP2, X2_len, XP2_len, Y = self.get_batch(
)
# print('shapes: X1:{}, XP1:{}, X2:{}, XPS:{}, Y:{}'.format(X1.shape, XP1.shape, X2.shape, XP2.shape, Y.shape))
if self.config['use_dpool']:
yield ({
'query':
X1,
'query_pos':
XP1,
'query_len':
X1_len,
'query_pos_len':
XP1_len,
'doc':
X2,
'doc_pos':
XP2,
'doc_len':
X2_len,
'doc_pos_len':
XP2_len,
'dpool_index':
DynamicMaxPooling.dynamic_pooling_index(
X1_len, X2_len, self.config['text1_maxlen'],
self.config['text2_maxlen']),
'dpool_pos_index':
DynamicMaxPooling.dynamic_pooling_index(
XP1_len, XP2_len, self.config['pos1_maxlen'],
self.config['pos2_maxlen'])
}, Y)
else:
yield ({
'query': X1,
'query_pos': XP1,
'query_len': X1_len,
'query_pos_len': XP1_len,
'doc': X2,
'doc_pos': XP2,
'doc_len': X2_len,
'doc_pos_len': XP2_len
}, Y)
def build(self):
query = Input(name='query', shape=(self.config['text1_maxlen'], ))
show_layer_info('Input', query)
doc = Input(name='doc', shape=(self.config['text2_maxlen'], ))
show_layer_info('Input', doc)
dpool_index = Input(name='dpool_index',
shape=[
self.config['text1_maxlen'],
self.config['text2_maxlen'], 3
],
dtype='int32')
show_layer_info('Input', dpool_index)
embedding = Embedding(self.config['vocab_size'],
self.config['embed_size'],
weights=[self.config['embed']],
trainable=self.embed_trainable)
q_embed = embedding(query)
show_layer_info('Embedding', q_embed)
d_embed = embedding(doc)
show_layer_info('Embedding', d_embed)
cross = Dot(axes=[2, 2], normalize=False)([q_embed, d_embed])
show_layer_info('Dot', cross)
cross_reshape = Reshape((self.config['text1_maxlen'],
self.config['text2_maxlen'], 1))(cross)
show_layer_info('Reshape', cross_reshape)
conv2d = Conv2D(self.config['kernel_count'],
self.config['kernel_size'],
padding='same',
activation='relu')
dpool = DynamicMaxPooling(self.config['dpool_size'][0],
self.config['dpool_size'][1])
conv1 = conv2d(cross_reshape)
show_layer_info('Conv2D', conv1)
pool1 = dpool([conv1, dpool_index])
show_layer_info('DynamicMaxPooling', pool1)
pool1_flat = Flatten()(pool1)
show_layer_info('Flatten', pool1_flat)
pool1_flat_drop = Dropout(rate=self.config['dropout_rate'])(pool1_flat)
show_layer_info('Dropout', pool1_flat_drop)
if self.config['target_mode'] == 'classification':
out_ = Dense(2, activation='softmax')(pool1_flat_drop)
elif self.config['target_mode'] in ['regression', 'ranking']:
out_ = Dense(1)(pool1_flat_drop)
show_layer_info('Dense', out_)
model = Model(inputs=[query, doc, dpool_index], outputs=out_)
return model
def build(self):
query = Input(name='query', shape=(self.config['text1_maxlen'], ))
show_layer_info('Input', query)
doc = Input(name='doc', shape=(self.config['text2_maxlen'], ))
show_layer_info('Input', doc)
dpool_index = Input(name='dpool_index',
shape=[
self.config['text1_maxlen'],
self.config['text2_maxlen'], 3
],
dtype='int32')
show_layer_info('Input', dpool_index)
embedding = Embedding(self.config['vocab_size'],
self.config['embed_size'],
weights=[self.config['embed']],
trainable=self.embed_trainable)
q_embed = embedding(query)
show_layer_info('Embedding', q_embed)
d_embed = embedding(doc)
show_layer_info('Embedding', d_embed)
# ########## compute attention weights for the query words: better then mvlstm alone
if self.config["text1_attention"]:
q_w = Dense(1,
kernel_initializer=self.initializer_gate,
use_bias=False)(
q_embed) # use_bias=False to simple combination
show_layer_info('Dense', q_w)
q_w = Lambda(lambda x: softmax(x, axis=1),
output_shape=(self.config['text1_maxlen'], ),
name="q_w")(q_w)
show_layer_info('Lambda-softmax', q_w)
# ########## add attention weights for Q_words
q_w_layer = Lambda(lambda x: K.repeat_elements(
q_w, rep=self.config['embed_size'], axis=2))(q_w)
show_layer_info('repeat', q_w_layer)
q_embed = Multiply()([q_w_layer, q_embed])
show_layer_info('Dot-qw', q_embed)
# ####################### attention text1
# ########## compute attention weights for the document words:
if self.config['text2_attention']:
d_w = Dense(1,
kernel_initializer=self.initializer_gate,
use_bias=False)(d_embed)
show_layer_info('Dense', d_w)
d_w = Lambda(lambda x: softmax(x, axis=1),
output_shape=(self.config['text2_maxlen'], ))(d_w)
show_layer_info('Lambda-softmax', d_w)
# ########## add attention weights for D_words
d_w_layer = Lambda(lambda x: K.repeat_elements(
d_w, rep=self.config['embed_size'], axis=2))(d_w)
d_embed = Multiply()([d_w_layer, d_embed])
show_layer_info('Dot-qw', d_embed)
# ####################### attention text2
cross = Dot(axes=[2, 2], normalize=False)([q_embed, d_embed])
show_layer_info('Dot', cross)
cross_reshape = Reshape((self.config['text1_maxlen'],
self.config['text2_maxlen'], 1))(cross)
show_layer_info('Reshape', cross_reshape)
conv2d = Conv2D(self.config['kernel_count'],
self.config['kernel_size'],
padding='same',
activation='relu')
dpool = DynamicMaxPooling(self.config['dpool_size'][0],
self.config['dpool_size'][1])
conv1 = conv2d(cross_reshape)
show_layer_info('Conv2D', conv1)
pool1 = dpool([conv1, dpool_index])
show_layer_info('DynamicMaxPooling', pool1)
pool1_flat = Flatten()(pool1)
show_layer_info('Flatten', pool1_flat)
pool1_flat_drop = Dropout(rate=self.config['dropout_rate'])(pool1_flat)
show_layer_info('Dropout', pool1_flat_drop)
if self.config['target_mode'] == 'classification':
out_ = Dense(2, activation='softmax')(pool1_flat_drop)
elif self.config['target_mode'] in ['regression', 'ranking']:
out_ = Dense(1)(pool1_flat_drop)
show_layer_info('Dense', out_)
model = Model(inputs=[query, doc, dpool_index], outputs=out_)
return model
def get_batch_generator(self):
for X1, X1_len, X2, X2_len, Y, ID_pairs, list_counts in self.get_batch():
if self.config['use_dpool']:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len, 'dpool_index': DynamicMaxPooling.dynamic_pooling_index(X1_len, X2_len, self.config['text1_maxlen'], self.config['text2_maxlen']), 'ID': ID_pairs, 'list_counts': list_counts}, Y)
else:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len, 'ID': ID_pairs, 'list_counts': list_counts}, Y)
def build(self):
query = Input(name='query', shape=(self.config['text1_maxlen'], ))
show_layer_info('Input', query)
doc = Input(name='doc', shape=(self.config['text2_maxlen'], ))
show_layer_info('Input', doc)
dpool_index = Input(name='dpool_index',
shape=[
self.config['text1_maxlen'],
self.config['text2_maxlen'], 3
],
dtype='int32')
show_layer_info('Input', dpool_index)
embedding = Embedding(self.config['vocab_size'],
self.config['embed_size'],
weights=[self.config['embed']],
trainable=self.embed_trainable)
q_embed = embedding(query)
show_layer_info('Embedding', q_embed)
d_embed = embedding(doc)
show_layer_info('Embedding', d_embed)
cross = Dot(axes=[2, 2], normalize=False)([q_embed, d_embed])
# def cal_binsum(cross, bin_num=20):
# shape = cross.get_shape()
# qnum = shape[1]
# mbinsum = np.zeros((qnum, bin_num), dtype=np.float32)
# for (i, j), v in np.ndenumerate(cross):
# if i >= qnum:
# break
# vid = int((v + 1.) / 2. * (bin_num - 1.))
# mbinsum[i][vid] += v
# # mhist += 1. # smooth is not needed for computing bin sum
# # mhist = np.log10(mhist) # not needed for computing bin sum
#
# return mbinsum.flatten()
#
# bins = Lambda(lambda x: cal_binsum(x))(cross)
# out1 = Dense(50)(bins)
show_layer_info('Dot', cross)
cross_reshape = Reshape((self.config['text1_maxlen'],
self.config['text2_maxlen'], 1))(cross)
show_layer_info('Reshape', cross_reshape)
conv2d = Conv2D(self.config['kernel_count'],
self.config['kernel_size'],
padding='same',
activation='relu')
dpool = DynamicMaxPooling(self.config['dpool_size'][0],
self.config['dpool_size'][1])
conv1 = conv2d(cross_reshape)
show_layer_info('Conv2D', conv1)
pool1 = dpool([conv1, dpool_index])
show_layer_info('DynamicMaxPooling', pool1)
pool1_flat = Flatten()(pool1)
show_layer_info('Flatten', pool1_flat)
pool1_flat_drop = Dropout(rate=self.config['dropout_rate'])(pool1_flat)
show_layer_info('Dropout', pool1_flat_drop)
if self.config['target_mode'] == 'classification':
out_ = Dense(2, activation='softmax')(pool1_flat_drop)
elif self.config['target_mode'] in ['regression', 'ranking']:
out_ = Dense(1)(pool1_flat_drop)
show_layer_info('Dense', out_)
model = Model(inputs=[query, doc, dpool_index], outputs=out_)
model.summary()
return model
def get_batch_generator(self):
while True:
sample = self.get_batch()
if not sample:
break
X1, X1_len, X2, X2_len, Y, ID_pairs = sample
if self.config['use_dpool']:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len, 'dpool_index': DynamicMaxPooling.dynamic_pooling_index(X1_len, X2_len, self.config['text1_maxlen'], self.config['text2_maxlen']), 'ID':ID_pairs}, Y)
else:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len, 'ID':ID_pairs}, Y)
def get_batch_generator(self):
while True:
X1, X1_len, X2, X2_len, Y = self.get_batch()
if self.config['use_dpool']:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len, 'dpool_index': DynamicMaxPooling.dynamic_pooling_index(X1_len, X2_len, self.config['text1_maxlen'], self.config['text2_maxlen'])}, Y)
else:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len}, Y)
def get_batch_generator(self):
if self.is_train:
while True:
X1, X1_len, X2, X2_len, Y, ID_pairs = self.get_batch()
if self.config['use_dpool']:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len, 'dpool_index': DynamicMaxPooling.dynamic_pooling_index(X1_len, X2_len, self.config['text1_maxlen'], self.config['text2_maxlen'])}, Y)
else:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len}, Y)
else:
while self.point + self.batch_size <= self.total_rel_num:
X1, X1_len, X2, X2_len, Y, ID_pairs = self.get_batch(randomly = False)
if self.config['use_dpool']:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len, 'dpool_index': DynamicMaxPooling.dynamic_pooling_index(X1_len, X2_len, self.config['text1_maxlen'], self.config['text2_maxlen']), 'ID':ID_pairs}, Y)
else:
yield ({'query': X1, 'query_len': X1_len, 'doc': X2, 'doc_len': X2_len, 'ID':ID_pairs}, Y)
