tensorflow2报错AttributeError: module ‘tensorflow‘ has no attribute ‘truncated_normal‘

淡淡的烟草味﹌ 2023-02-28 04:28 33阅读 0赞

## tensorflow2报错AttributeError: module ‘tensorflow’ has no attribute ‘truncated\_normal’ ##

#### 代码 ####

tn = tf.truncated_normal([prev_num_nodes, num_nodes], stddev=w_stddev)

#### 报错信息 ####

---------------------------------------------------------------------------
    AttributeError                            Traceback (most recent call last)
    <ipython-input-14-05dbe3f468af> in <module>()
          5 
          6 # new_alex = TransferedAlexNet(old_alex_graph, num_tgt_classes=2)
    ----> 7 new_alex = TransferedAlexNet(old_alex_graph)
          8 print(new_alex)
          9 inputs   = old_alex_graph.get_tensor_by_name('inputs/images:0')
    
    <ipython-input-12-d49ea2971fb0> in __init__(s, init_graph)
         21             with tf.name_scope('new_top_layer'):
         22                 # the old fc8 (which we are replacing) had 1000 nodes for 1000 classes
    ---> 23                 s.logits = helpers_07.fully_connected_xavier_relu_layer(orig_fc7)
         24                 print( s.logits)
         25 
    
     in fully_connected_xavier_relu_layer(incoming_layer, num_nodes, b_val, keep_prob, name)
         84                                  activation_fn = tf.nn.relu,
         85                                  keep_prob=keep_prob,
    ---> 86                                  name=name)
         87 
         88 if __name__ == "__main__":
    
    in fully_connected_layer(incoming_layer, num_nodes, w_stddev, b_val, activation_fn, keep_prob, name)
         56         # FIXME:  consider stddev=np.sqrt(2.0 / prev_num_nodes)
         57         #         [fancy initialization]
    ---> 58         tn = tf.truncated_normal([prev_num_nodes, num_nodes], stddev=w_stddev)
         59         W = tf.Variable(tn, name='W')
         60         const = tf.constant(b_val, shape=[num_nodes])
    
    AttributeError: module 'tensorflow' has no attribute 'truncated_normal'

### 解决方法 ###

tf.random.truncated_normal([prev_num_nodes, num_nodes], stddev=w_stddev)