扩大CNN的输出

Question

我想将我的网络输出从11个输出扩展到12个输出，我恢复了之前已在11个输出上重新训练的检查点。我从here找到了答案。利用这一点，我发现如何改变输出图层的形状，将其展开以适应另一排权重，但我不知道是否我正确初始化了重量和偏差。实际上，我没有收到任何编译或运行时错误，但测试精度从95％下降到9％。代码中的某处可能有问题。下面的代码：

w_b_not = { 
     'weight_4': tf.Variable(tf.random_normal([num_hidden, num_labels], stddev=0.1)), 
     'bias_4' : tf.Variable(tf.constant(1.0, shape=[num_labels])),} 

w_b = { 
     'wc1_0': tf.Variable(tf.random_normal([patch_size_1, patch_size_1, num_channels, depth],stddev=0.1)), 
     ..... 
     'bc1_0' : tf.Variable(tf.zeros([depth]))} 

    .... #here is the networks model 

num_steps = 1001 
with tf.Session(graph=graph) as sess: 
    ckpt = ('path_of_checkpoint.ckpt') 
    if os.path.isfile(ckpt) : 
    layer6_weights = tf.Variable(tf.random_normal([num_hidden, num_labels], stddev=0.1)) 
    layer6_biases = tf.Variable(tf.constant(1.0, shape=[num_labels])) 

    n_w_b = { 
     'new_layer_weights' : tf.concat(0,[w_b_not['weight_4'], layer6_weights]), 
     'new_layer_biases' : tf.concat(0,[w_b_not['bias_4'], layer6_biases])} 
    resize_var_1 = tf.assign(w_b_not['weight_4'], n_w_b['new_layer_weights'], validate_shape=False) 
    resize_var_2 = tf.assign(w_b_not['bias_4'], n_w_b['new_layer_biases'], validate_shape=False) 
    logits = tf.get_collection('logits')[0] 
    w_b_new_saver = tf.train.Saver() 
    init_op = tf.initialize_all_variables()   
    w_b_saver.restore(sess, ckpt) 
    print("restore complete") 
    for step in xrange(num_steps): 
     sess.run(init_op) 
    print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval() , test_labels, force = False)) 

删除后for循环我得到这个错误：

Traceback (most recent call last): 
File "/home/owner/tensorflow/tensorflow/models/image/mnist/new_dataset/Nets.py", line 237, in <module> 
print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval() , test_labels, force = False)) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 502, in eval 
return _eval_using_default_session(self, feed_dict, self.graph, session) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 3334, in _eval_using_default_session 
return session.run(tensors, feed_dict) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 340, in run 
run_metadata_ptr) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 564, in _run 
feed_dict_string, options, run_metadata) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 637, in _do_run 
target_list, options, run_metadata) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 659, in _do_call 
e.code) 
FailedPreconditionError: Attempting to use uninitialized value Variable 
[[Node: Variable/read = Identity[T=DT_FLOAT, _class=["loc:@Variable"], _device="/job:localhost/replica:0/task:0/gpu:0"](Variable)]] 
[[Node: Softmax_2/_19 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:localhost/replica:0/task:0/gpu:0", send_device_incarnation=1, tensor_name="edge_715_Softmax_2", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"]()]] 
Caused by op u'Variable/read', defined at: 
File "<string>", line 1, in <module> 
File "/usr/lib/python2.7/idlelib/run.py", line 116, in main 
ret = method(*args, **kwargs) 
File "/usr/lib/python2.7/idlelib/run.py", line 324, in runcode 
exec code in self.locals 
File "/home/owner/tensorflow/tensorflow/models/image/mnist/new_dataset/Nets.py", line 155, in <module> 
'weight_4': tf.Variable(tf.random_normal([num_hidden, num_labels], stddev=0.1)), 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/variables.py", line 206, in __init__ 
dtype=dtype) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/variables.py", line 275, in _init_from_args 
self._snapshot = array_ops.identity(self._variable, name="read") 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/gen_array_ops.py", line 609, in identity 
return _op_def_lib.apply_op("Identity", input=input, name=name) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/op_def_library.py", line 655, in apply_op 
op_def=op_def) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 2154, in create_op 
original_op=self._default_original_op, op_def=op_def) 
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 1154, in __init__ 
self._traceback = _extract_stack() 

Answer 1

更改此

w_b_new_saver = tf.train.Saver() 
init_op = tf.initialize_all_variables()   
w_b_saver.restore(sess, ckpt) 
print("restore complete") 
for step in xrange(num_steps): 
    sess.run(init_op) 
print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval() , test_labels, force = False)) 

到：

w_b_new_saver = tf.train.Saver()  
w_b_saver.restore(sess, ckpt) 
print("restore complete") 
sess.run(tf.initialize_variables(list(tf.get_variable(name) for name in sess.run(tf.report_uninitialized_variables(tf.all_variables()))))) 
# Only initialize the unitialize variables 
# That really long line will go through the variables and check which ones are not initialized 
# If you already know which ones aren't initialized just pass those in directly 
w_b_new_saver2 = tf.train.Saver() # Now you can make a new saver 
# in case you want to save this changed model with new weights 
print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval() , test_labels, force = False)) 

您重新初始化变量这会导致你准确性下降回落到9％，因为你正在重置所有的权重。