background:
I do not know if you have no such worries: build a good model in use Tensorflow debugging process, often encounter some problems, then spent a lot of time to this issue is resolved, after a period of time, there was the same problem, but how do not remember how to solve before, and the results they have to do it again.
In order to prevent this from happening again, especially in this record, for later investigation.
text:
Questions are listed below:
Question 1:
Did you mean to set reuse=True or reuse =tf.AUTO_REUSE in Varscope?
Solution:
增加一行:tf.get_variable_scope_scope().reuse_variables() in compute_loss / accuracy
Question 2:
Fetch argument32. has invalid <type> cannot convert a float32 into a Tensor or Operation
Solution:
Variable statement can not be the same before and after the run.