An "InvalidArgumentError" BUG occurred during the reinforcement learning stage
Opened this issue · 2 comments
Hi!
I have run your code with TensorFlow1.4-cpu, TensorFlow1.5-gpu and TensorFlow1.12-gpu
(I made modification on original code like hps.hidden_dim -> hps.hidden_dim.value so that it can be run with TensorFlow version higher than 1.4 )
But all of them occurred the same "InvalidArgumentError" during reinforcement learning stage. The complete error message is in https://pastebin.com/z5gp2srh. By the way, to make thing easier, i reduce the the number of epochs of each pre-traning but i sure this is not the reason of the error.
According to some relative information from internet, this error seems caused by inappropriate magnitudes of some reinforcement learning parameters, but I cannot find the exact one. It would be so nice if you would like to help me with this problem.
@WaterFriend Hello there, I am running tensorflow based on CPU 1.4.1 linux version (or even 1.4.0 for windows), and I am facing the same error now. Did you find any work around for this issue? Thanks.
@WaterFriend Hello there, I am running tensorflow based on CPU 1.4.1 linux version (or even 1.4.0 for windows), and I am facing the same error now. Did you find any work around for this issue? Thanks.
Actually, until now I still cannot find the solution to that error. What I did to deal with the dilemma was building up the environment strictly base on the requirement. This is the most secure way to ensure everything can run smoothly.