File size: 1,833 Bytes
6077218
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# import tensorflow as tf


# def optimizer(learner,loss,learning_rate,momentum=0.9):
#     optimizer=None
#     if learner.lower() == "adagrad": 
#         optimizer = tf.train.AdagradOptimizer(learning_rate=learning_rate,\
#                      initial_accumulator_value=1e-8).minimize(loss)
#     elif learner.lower() == "rmsprop":
#         optimizer = tf.train.RMSPropOptimizer(learning_rate).minimize(loss)
#     elif learner.lower() == "adam":
#         optimizer = tf.train.AdamOptimizer(learning_rate).minimize(loss)
#     elif learner.lower() == "gd" :
#         optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss)  
#     elif learner.lower() == "momentum" :
#         optimizer = tf.train.MomentumOptimizer(learning_rate,momentum).minimize(loss)  
#     else:
#         raise ValueError("please select a suitable optimizer")  
#     return optimizer

# def pairwise_loss(loss_function,y,margin=1):
#     loss=None
#     if loss_function.lower() == "bpr":
#         loss = -tf.reduce_sum(tf.log_sigmoid(y))
#     elif loss_function.lower() == "hinge":
#         loss = tf.reduce_sum(tf.maximum(y+margin, 0))
#     elif loss_function.lower() == "square":  
#         loss = tf.reduce_sum(tf.square(1-y))   
#     else:
#         raise Exception("please choose a suitable loss function")
#     return loss

# def pointwise_loss(loss_function,y_rea,y_pre):
#     loss=None
#     if loss_function.lower() == "cross_entropy":
#         loss = tf.losses.sigmoid_cross_entropy(y_rea,y_pre)
# #         loss = - tf.reduce_sum(
# #             y_rea * tf.log(y_pre) + (1 - y_rea) * tf.log(1 - y_pre)) 
#     elif loss_function.lower() == "square":  
#         loss = tf.reduce_sum(tf.square(y_rea-y_pre))  
#     else:
#         raise Exception("please choose a suitable loss function") 
#     return loss