在TensorFlow中构建模型时如何解决ValueError？

Question

When I run the code written on https://github.com/MG2033/ShuffleNet , just by running python main.py --config config/test.json , I get the following error when ‌buling the model:

Building the model...
Traceback (most recent call last):
File "main.py", line 69, in <module>
main()
File "main.py", line 35, in main
model = ShuffleNet(config_args)
File "/home/farzaneh/PycharmProjects/shufflenet code/ShuffleNet-master/model.py", line 29, in __init__ 
self.__build()
File "/home/farzaneh/PycharmProjects/shufflenet code/ShuffleNet-master/  
model.py", line 134, in __build
self.__init_output()
File "/home/farzaneh/PycharmProjects/shufflenet code/ShuffleNet-master  /model.py", line 91, in __init_output
self.check_op = tf.add_check_numerics_ops()
File "/home/farzaneh/anaconda3/lib/python3.6/site-packages/tensorflow /python/ops/numerics.py", line 75, in add_check_numerics_ops
raise ValueError("`tf.add_check_numerics_ops() is not compatible "
ValueError: `tf.add_check_numerics_ops() is not compatible with TensorFlow    control flow operations such as `tf.cond()` or `tf.while_loop()`.

How can I fix it?

Answer 1

According to the readme of shuffle.NET, you need to provide a path to the config file as argument, as following: python main.py --config config/test.json . And there should be a config.

EDIT: After I looked into it:
Apparently, the error is right and add_check_numerics_ops is incompatible with while_loops and stuff. In the code you can also see that this line was just added for debugging to handle nans, which normally should not occur. Therefore I guess you can just comment out this part. Finally I found that there is this issue on the project too, which gives me the feeling that there is really an issue with the code.