问题描述
我在使用Tensorflow（2.3）执行自定义操作时编写的代码段遇到了问题。该代码通常可以正常工作，但有时即使在其他具有相同输入的执行中也可以正常工作，有时也会引发意外的错误和异常。
我试图解决该问题，并且我几乎确信这是一个评估依赖问题。我试图添加一些依赖项控件，但这没有用。我为一点点冗长的代码道歉，实际上我无法在一个较小的示例中重现该问题。下面是我的代码：
import tensorflow.compat.v1 as tf
tf.compat.v1.disable_eager_execution()
tf.disable_v2_behavior()
myTensor_values = tf.placeholder(dtype=tf.float32)
myTensor_l2_splits = tf.placeholder(dtype=tf.int32)
myTensor_l1_splits = tf.placeholder(dtype=tf.int32)
def innerloop_processing(begin_index,end_index,input1) :
innerloop_counter = begin_index
ta = tf.TensorArray(tf.float32,size=0,dynamic_size=True,clear_after_read=False,infer_shape=False )
def innerloop_body(counter,begin_index,input1,ta) :
inner_being_index = input1[1][counter]
inner_end_index = input1[1][counter+1]
row = tf.slice(input1[0],[inner_being_index],[inner_end_index-inner_being_index])
ta = ta.write(counter-begin_index,row)
counter = counter + 1
return counter,ta
def innerloop_cond(counter,ta ) :
return input1[1][counter] < input1[1][end_index] -1 #stop at the next pointer of the l2_splits
results = tf.while_loop(innerloop_cond,innerloop_body,[innerloop_counter,ta] )
print_resutls = tf.print("this is the component result :",results[4].stack())
return results[4].stack()
def generateL1Tensor_writeback(start_offest,step,num):
counter=tf.constant(0,tf.int32)
values = tf.TensorArray(tf.int32,infer_shape=False )
def cond(values,start_offest,num,counter) :
return counter*step <= num*step
def body(values,counter) :
values = values.write(counter,[(counter*step)+start_offest])
counter = counter+1
return values,counter
final_values,_,_ = tf.while_loop(cond,body,[values,counter])
final = final_values.concat()
#print_line = tf.print(" xxxxx This is the is the split : ",final)
return final
def multiply2n_ragged(tensor1,tensor2) :
#this function multiplies two ragged tesnsors of rank 2 . the most outer ranks of the two tensros must be equal .
#setting variables and constats
outerloop_counter = tf.constant(0,dtype=tf.int32)
carry_on = tf.constant(0,dtype=tf.int32)
taValues = tf.TensorArray(tf.float32,infer_shape=False )
taL2Splits = tf.TensorArray(tf.int32,infer_shape=False )
taL1Splits = tf.TensorArray(tf.int32,infer_shape=False )
taL1Splits = taL1Splits.write(0,[0]) ## required intialization for L1 split only
innerloop_processing_graphed = tf.function(innerloop_processing)
generateL1Tensor_writeback_graphed = tf.function(generateL1Tensor_writeback)
def outerloop_cond(counter,input2,taValues,taL2Splits,taL1Splits,carry_on ) :
value = tf.shape(input1[2])[0]-1
return counter < value ## this is the length of the outermost dimision,stop of this
def outloop_body(counter,carry_on) :
l1_comp_begin = input1[2][counter] ## this is begin position of the current row in the outer split ( ie. the ith value in the outer row split tensor )
l1_comp_end = input1[2][counter+1] ## this is end position of the current row in the outer split (ie. the ith + 1 value in the outer row split tensor)
l1_comp2_begin = input2[2][counter] ## we do the same for the second components
l1_comp2_end = input2[2][counter+1] ## we do the same for the second components
comp = innerloop_processing_graphed(l1_comp_begin,l1_comp_end,input1 ) ## Now retrive the data to be procesed for the selected rows from vector1
comp2 =innerloop_processing_graphed(l1_comp2_begin,l1_comp2_end,input2 ) ## do the same for vector 2
comp2 = tf.transpose(comp2) ### desired operation
multiply =tf.matmul(comp,comp2) #### This is the desired operation
myshape= tf.shape(multiply) ## calculate the shape of the result in order to prepare to write the result in a ragged tensor format.
offset = tf.cond( taValues.size() >0,lambda: tf.shape(taValues.concat())[0],lambda : [0]) ### this is a hack,TensorArray.concat returns an error if the array is empty. Thus we check before calling this.
l2v = generateL1Tensor_writeback_graphed(offset,myshape[1],myshape[0]) # generate the inner row split of the result for the current element
taL2Splits=taL2Splits.write(counter,l2v) # write back the inner rowlplit to a TensorArray
taValues=taValues.write(counter,tf.reshape(multiply,[-1])) # wirte back the actual ragged tensor elemnts in a another TensorArray
carry_on=carry_on+myshape[0] ## required to calculate the outer row splite
taL1Splits=taL1Splits.write(counter+1,[carry_on]) ## This is the outmost row split.
counter = counter+1
return counter,carry_on
outerloop_finalcounter,ta1,ta2,ta3,_ = tf.while_loop(outerloop_cond,outloop_body,[outerloop_counter,tensor1,tensor2,carry_on])
uinquie_ta2,_ = tf.unique(ta2.concat()) # this is required since some values might be duplicate in the row split itself
final_values = ta1.concat(),uinquie_ta2,ta3.concat()
return final_values
t = myTensor_values,myTensor_l2_splits,myTensor_l1_splits
oo =multiply2n_ragged(t,t)
new_oo = multiply2n_ragged(oo,oo)
sess = tf.Session(config=tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True)))
sess.run(tf.global_variables_initializer())
vals =np.array([1.0,2.2,1.1,4.0,5.0,6.0,7.0,8.0,9.0,10.0,11.0,1.1 ])
l2_splits = np.array([0,3,6,9,12,15])
l1_splits = np.array([0,2,5 ])
re = sess.run([new_oo ],Feed_dict={myTensor_values:vals,myTensor_l1_splits:l1_splits,myTensor_l2_splits:l2_splits } )
print(re)
正如我所说，该代码可以多次正常运行，但是对于相同的输入，有时会产生以下错误。我得到的一些不同错误的堆栈痕迹：
this is the component result : [[1 2.2 1.1]
[4 5 1.1]]
this is the component result : [[1 2.2 1.1]
[4 5 1.1]]
this is the component result : [[6 7 1.1]
[8 9 1.1]
[10 11 1.1]]
this is the component result : [[6 7 1.1]
[8 9 1.1]
[10 11 1.1]]
this is the component result : [[7.05 16.21]
[16.21 42.21]]
this is the component result : [[7.05 16.21]
[16.21 42.21]]
---------------------------------------------------------------------------
InvalidArgumentError Traceback (most recent call last)
C:\ProgramData\Anaconda3\envs\AutoEncoder\lib\site-packages\tensorflow\python\client\session.py in _do_call(self,fn,*args)
1364 try:
-> 1365 return fn(*args)
1366 except errors.OpError as e:
C:\ProgramData\Anaconda3\envs\AutoEncoder\lib\site-packages\tensorflow\python\client\session.py in _run_fn(Feed_dict,fetch_list,target_list,options,run_Metadata)
1349 return self._call_tf_sessionrun(options,Feed_dict,-> 1350 target_list,run_Metadata)
1351
C:\ProgramData\Anaconda3\envs\AutoEncoder\lib\site-packages\tensorflow\python\client\session.py in _call_tf_sessionrun(self,run_Metadata)
1442 fetch_list,-> 1443 run_Metadata)
1444
InvalidArgumentError: {{function_node __inference_innerloop_processing_13658}} {{function_node __inference_innerloop_processing_13658}} Expected size[0] in [0,0],but got 3
[[{{node while/body/_1/while/Slice}}]]
[[while_33/StatefulPartitionedCall_1]]
During handling of the above exception,another exception occurred:
InvalidArgumentError Traceback (most recent call last)
<ipython-input-18-238a2ce9a03a> in <module>
94 l2_splits = np.array([0,15])
95 l1_splits = np.array([0,5 ])
---> 96 re = sess.run([new_oo ],myTensor_l2_splits:l2_splits } )
97 print(re)
C:\ProgramData\Anaconda3\envs\AutoEncoder\lib\site-packages\tensorflow\python\client\session.py in run(self,fetches,run_Metadata)
956 try:
957 result = self._run(None,options_ptr,--> 958 run_Metadata_ptr)
959 if run_Metadata:
960 proto_data = tf_session.TF_GetBuffer(run_Metadata_ptr)
C:\ProgramData\Anaconda3\envs\AutoEncoder\lib\site-packages\tensorflow\python\client\session.py in _run(self,handle,run_Metadata)
1179 if final_fetches or final_targets or (handle and Feed_dict_tensor):
1180 results = self._do_run(handle,final_targets,final_fetches,-> 1181 Feed_dict_tensor,run_Metadata)
1182 else:
1183 results = []
C:\ProgramData\Anaconda3\envs\AutoEncoder\lib\site-packages\tensorflow\python\client\session.py in _do_run(self,run_Metadata)
1357 if handle is None:
1358 return self._do_call(_run_fn,Feeds,targets,-> 1359 run_Metadata)
1360 else:
1361 return self._do_call(_prun_fn,fetches)
C:\ProgramData\Anaconda3\envs\AutoEncoder\lib\site-packages\tensorflow\python\client\session.py in _do_call(self,*args)
1382 '\nsession_config.graph_options.rewrite_options.'
1383 'disable_Meta_optimizer = True')
-> 1384 raise type(e)(node_def,op,message)
1385
1386 def _extend_graph(self):
InvalidArgumentError: Expected size[0] in [0,but got 3
[[{{node while/body/_1/while/Slice}}]]
[[while_33/StatefulPartitionedCall_1]]
以及以下错误：
CancelledError Traceback (most recent call last)
C:\ProgramData\Anaconda3\envs\AutoEncoder\lib\site-packages\tensorflow\python\client\session.py in _do_call(self,-> 1443 run_Metadata)
1444
CancelledError: {{function_node __inference_innerloop_processing_11240}} {{function_node __inference_innerloop_processing_11240}} [_Derived_]Loop execution was cancelled.
[[{{node while/LoopCond/_20}}]]
[[while_27/StatefulPartitionedCall_1]]
During handling of the above exception,another exception occurred:
CancelledError Traceback (most recent call last)
<ipython-input-15-238a2ce9a03a> in <module>
94 l2_splits = np.array([0,message)
1385
1386 def _extend_graph(self):
CancelledError: [_Derived_]Loop execution was cancelled.
[[{{node while/LoopCond/_20}}]]
[[while_27/StatefulPartitionedCall_1]]
我相信所有错误都将抛出innerloop_processing中。我还在Tensorflow github here中打开了一个问题。
解决方法
问题似乎出自 tf.Cond，幸运的是在 tensorflow2 中重新实现了这一点。从而删除调用：
tf.disable_v2_behavior()
解决问题。