Open lixin4ever opened 6 years ago
@lixin4ever I tried several versions of Theano, but all of them didn't work well. At last, Theano==0.8.0 with CPU as a compromise proposal can get the results. FYI. However, GPU couldn't work caused by the version CUDnn, I think.
@ArrowLuo Thanks for your notification. Actually, I have figured it out by changing some part of the source code.
Has anyone made it work with GPU support yet?
@lixin4ever I have tried this code with several versions of Theano, but I am unable to run it successfully. Can you help me out to run this code by providing your edited source code? Thanks in anticipation
@JamesRiddle I made it run with python2 theano 0.8.2. If that doesn't work for you, let me know.
@JamesRiddle I am ok for providing the modified code but I think you can follow the idea provided by @hnstr first. If that doesn't work, we can discuss here so that the following users can know what is wrong with the original code.
Yes, the code should be running on theano 0.8.2. That's the version I used for the implementation.
On Thu, Jun 28, 2018 at 7:35 PM, LI XIN notifications@github.com wrote:
@JamesRiddle https://github.com/JamesRiddle I am ok for providing the modified code but I think you can follow the idea provided by @hnstr https://github.com/hnstr first. If that doesn't work, we can discuss here so that the following users can know what is wrong with the original code.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/happywwy/Coupled-Multi-layer-Attentions/issues/1#issuecomment-401005128, or mute the thread https://github.com/notifications/unsubscribe-auth/AGKJI1F12SmXBet8zZUnNo_Z1dkMsDGjks5uBL-KgaJpZM4RQtVM .
@JamesRiddle I am ok for providing the modified code but I think you can follow the idea provided by @hnstr first. If that doesn't work, we can discuss here so that the following users can know what is wrong with the original code.
Hi, lixin. Can you share your modified code here? I also failed to run the original code, even under the python2+theano0.8.2. It will be very kind of you to let us know what is wrong with the original code.
Hi Wenya, Thanks for releasing the code. It seems that the code can not run normally and the problem lies in the recurrence function of GRU (i.e., the function "gru"). Can you tell me the theano version you use in the experiment?