-
Notifications
You must be signed in to change notification settings - Fork 0
/
tp_amtl_unconstrained.py
219 lines (173 loc) · 11.6 KB
/
tp_amtl_unconstrained.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
import tensorflow as tf
import numpy as np
import pdb
class TP_AMTL_UNCONSTRAINED(object):
def __init__(self, config):
for name in config.__dict__:
setattr(self,name,getattr(config,name))
self.x = tf.placeholder(shape=[None, config.num_steps, config.num_features], dtype=tf.float32, name='data')
self.y = tf.placeholder(shape=[None, config.num_tasks], dtype=tf.float32, name='labels')
self.num_samples_ph = tf.placeholder(dtype=tf.int32,name='num_samples')
self.train = tf.placeholder(dtype=tf.bool,name='train')
self.keep_prob = 0.7
self.global_step = tf.Variable(0, trainable=False)
self.lr_decay = tf.train.exponential_decay(self.lr, self.global_step,
10000, 0.8, staircase=False)
self.build_model()
def output(self, task_id, embed, beta_output):
with tf.variable_scope("task_"+str(task_id)+'/output'):
beta_att = tf.layers.dense(beta_output,self.num_hidden,activation=tf.nn.tanh,use_bias=True,name='beta_att')
c_i = tf.reduce_mean(beta_att * embed, 1)
logits = tf.layers.dense(c_i,1,activation=None,use_bias=True,name='output_layer')
self.beta_att_each.append(beta_att)
return logits
def transfer(self,information_source,information_target,V,to_task,from_task):
with tf.variable_scope('transfer_to%d_from%d'%(to_task,from_task),reuse=tf.AUTO_REUSE):
helper = np.zeros([self.num_steps,self.num_steps])
for i in range(self.num_steps):
for j in range(self.num_steps):
helper[i][j] = 1.0
helper = tf.convert_to_tensor(helper,dtype=tf.float32)
information_target = tf.expand_dims(information_target,2)
information_target = tf.tile(information_target,[1,1,self.num_steps,1])
information_source = tf.expand_dims(information_source,1)
information_source = tf.tile(information_source,[1,self.num_steps,1,1])
information = tf.concat([information_target,information_source],3)
att = tf.layers.dense(information,self.num_hidden,activation=tf.nn.relu,use_bias=True)
att = tf.layers.dense(att,self.num_hidden,activation=tf.nn.relu,use_bias=True)
att = tf.layers.dense(att,1,activation=tf.nn.softplus,use_bias=True)
att = tf.squeeze(att,axis=[3])
att = att * helper
self.att_each[to_task,from_task].append(att)
with tf.variable_scope('transfer_from%d'%(from_task),reuse=tf.AUTO_REUSE):
transformed_V = tf.layers.dense(V,self.num_hidden,activation=tf.nn.leaky_relu,use_bias=True,name="layer1")
transformed_V = transformed_V / tf.norm(transformed_V,axis=2,keepdims=True)
transfer_amount = tf.matmul(att,transformed_V)
with tf.variable_scope('transfer_to%d'%(to_task),reuse=tf.AUTO_REUSE):
transfer_amount = tf.layers.dense(transfer_amount,self.num_hidden,activation=tf.nn.leaky_relu,use_bias=True,name="layer1")
return transfer_amount
def build_model(self, use_lstm=True):
print('Start building model')
self.loss_each = []
self.preds_each = []
self.beta_att_each = []
self.beta_main_each = []
loss_task = 0
self.att_each = {}
self.test = []
self.KL=[[0 for _ in range(self.num_tasks)] for i in range(self.num_tasks)]
self.att_loc = {}
self.att_scale = {}
self.att_each = {(i,j):[] for i in range(self.num_tasks) for j in range(self.num_tasks)}
self.att_each_before = {}
self.z_in = {}
self.test = {}
self.beta_output = []
self.helper = tf.contrib.distributions.fill_triangular(tf.ones([int(self.num_steps*(self.num_steps+1))/2]))
with tf.variable_scope("base"):
with tf.variable_scope("embed"):
embed = tf.layers.dense(self.x,self.num_hidden,activation=None,use_bias=False)
with tf.variable_scope("rnn"):
def single_cell():
lstm_cell = tf.contrib.rnn.LSTMCell(self.num_hidden)
return tf.contrib.rnn.DropoutWrapper(cell=lstm_cell,
output_keep_prob=self.keep_prob,
variational_recurrent=True,
dtype=tf.float32
)
cell = tf.contrib.rnn.MultiRNNCell([single_cell() for _ in range(self.num_shared_rnn_layers)])
r, _ = tf.nn.dynamic_rnn(cell,
embed,
dtype=tf.float32)
for task_id in range(self.num_tasks):
f = tf.reshape(r,[-1,self.num_hidden])
with tf.variable_scope('task_%d/ffw'%(task_id)):
for i in range(self.num_layers-1):
W = tf.get_variable('weight_'+str(i),[self.num_hidden,self.num_hidden])
B = tf.get_variable('bias_'+str(i),[self.num_hidden])
W = tf.nn.dropout(W,keep_prob=self.keep_prob)
f = tf.nn.leaky_relu(tf.matmul(f,W)+B)
W_loc1 = tf.get_variable('weight_loc1',[self.num_hidden,self.num_hidden])
B_loc1 = tf.get_variable('bias_loc1',[self.num_hidden])
W_loc1 = tf.nn.dropout(W_loc1,keep_prob=self.keep_prob)
beta_loc = tf.nn.leaky_relu(tf.matmul(f,W_loc1)+B_loc1)
W_loc2 = tf.get_variable('weight_loc2',[self.num_hidden,self.num_hidden])
B_loc2 = tf.get_variable('bias_loc2',[self.num_hidden])
W_loc2 = tf.nn.dropout(W_loc2,keep_prob=self.keep_prob)
beta_loc = tf.nn.leaky_relu(tf.matmul(beta_loc,W_loc2)+B_loc2)
beta_loc = tf.reshape(beta_loc,[-1,self.num_steps,self.num_hidden])
W_scale1 = tf.get_variable('weight_scale1',[self.num_hidden,self.num_hidden])
B_scale1 = tf.get_variable('bias_scale1',[self.num_hidden])
W_scale1 = tf.nn.dropout(W_scale1,keep_prob=self.keep_prob)
beta_scale = tf.nn.leaky_relu(tf.matmul(f,W_scale1)+B_scale1)
W_scale2 = tf.get_variable('weight_scale2',[self.num_hidden,self.num_hidden])
B_scale2 = tf.get_variable('bias_scale2',[self.num_hidden])
W_scale2 = tf.nn.dropout(W_scale2,keep_prob=self.keep_prob)
beta_scale = tf.nn.leaky_relu(tf.matmul(beta_scale,W_scale2)+B_scale2)
beta_scale = tf.reshape(beta_scale,[-1,self.num_steps,self.num_hidden])
beta_output = tf.distributions.Normal(beta_loc,beta_scale).sample()
self.beta_output.append(beta_output)
self.variance = []
for task_id in range(self.num_tasks):
beta_outputs = []
for s in range(self.num_training_samples):
with tf.variable_scope("rnn",reuse=True):
r, _ = tf.nn.dynamic_rnn(cell,
embed,
dtype=tf.float32)
f = tf.reshape(r,[-1,self.num_hidden])
with tf.variable_scope('task_%d/ffw'%(task_id),reuse=True):
for i in range(self.num_layers-1):
W = tf.get_variable('weight_'+str(i),[self.num_hidden,self.num_hidden])
B = tf.get_variable('bias_'+str(i),[self.num_hidden])
W = tf.nn.dropout(W,keep_prob=self.keep_prob)
f = tf.nn.leaky_relu(tf.matmul(f,W)+B)
W_loc1 = tf.get_variable('weight_loc1',[self.num_hidden,self.num_hidden])
B_loc1 = tf.get_variable('bias_loc1',[self.num_hidden])
W_loc1 = tf.nn.dropout(W_loc1,keep_prob=self.keep_prob)
beta_loc = tf.nn.leaky_relu(tf.matmul(f,W_loc1)+B_loc1)
W_loc2 = tf.get_variable('weight_loc2',[self.num_hidden,self.num_hidden])
B_loc2 = tf.get_variable('bias_loc2',[self.num_hidden])
W_loc2 = tf.nn.dropout(W_loc2,keep_prob=self.keep_prob)
beta_loc = tf.nn.leaky_relu(tf.matmul(beta_loc,W_loc2)+B_loc2)
beta_loc = tf.reshape(beta_loc,[-1,self.num_steps,self.num_hidden])
W_scale1 = tf.get_variable('weight_scale1',[self.num_hidden,self.num_hidden])
B_scale1 = tf.get_variable('bias_scale1',[self.num_hidden])
W_scale1 = tf.nn.dropout(W_scale1,keep_prob=self.keep_prob)
beta_scale = tf.nn.leaky_relu(tf.matmul(f,W_scale1)+B_scale1)
W_scale2 = tf.get_variable('weight_scale2',[self.num_hidden,self.num_hidden])
B_scale2 = tf.get_variable('bias_scale2',[self.num_hidden])
W_scale2 = tf.nn.dropout(W_scale2,keep_prob=self.keep_prob)
beta_scale = tf.nn.leaky_relu(tf.matmul(beta_scale,W_scale2)+B_scale2)
beta_scale = tf.reshape(beta_scale,[-1,self.num_steps,self.num_hidden])
beta_output = tf.stop_gradient(tf.distributions.Normal(beta_loc,beta_scale).sample())
beta_outputs.append(tf.expand_dims(beta_output,0))
_,variance = tf.nn.moments(tf.concat(beta_outputs,0), axes=[0])
self.variance.append(variance)
# transfer
self.beta_output_transfer = {}
for task_id in range(self.num_tasks):
beta_output_comb = self.beta_output[task_id]
for source_id in range(self.num_tasks):
if source_id != task_id:
source_knowledge = tf.stop_gradient(self.beta_output[source_id])
else:
source_knowledge = self.beta_output[source_id]
information_source = tf.concat([self.beta_output[source_id],self.variance[source_id]],2)
information_target = tf.concat([self.beta_output[task_id],self.variance[task_id]],2)
beta_output_transfer = self.transfer(tf.stop_gradient(information_source),tf.stop_gradient(information_target),source_knowledge,task_id,source_id)
beta_output_comb = beta_output_comb + beta_output_transfer
self.beta_output_transfer[source_id,task_id] = beta_output_transfer
# attention_sum
logits = self.output(task_id, embed, beta_output_comb)
preds = tf.nn.sigmoid(logits)
loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=logits, labels=self.y[:,task_id:task_id+1]))
self.loss_each.append(loss)
self.preds_each.append(preds)
loss_task += loss
l2_losses = [tf.nn.l2_loss(v) for v in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='base') if ('kernel' in v.name or 'weight' in v.name)]
loss_l2 = self.l2_coeff*tf.add_n(l2_losses)
self.loss_sum = loss_task + loss_l2
self.loss_all = {'loss_task':loss_task, 'loss_l2': loss_l2}
self.optim = tf.train.AdamOptimizer(self.lr).minimize(self.loss_sum)
print ('Model built')