incorporating components into keras-contrib

[ahundt]

Hey, I liked the clean way you implemented some of this code and since it is MIT licensed I was planning on merging a couple things in keras-contrib with attribution, but wanted to let you know!

Do you have any numbers for the kind of training results you got or idiosyncrasies I should know about?

[theduynguyen]

Hi Andrew, this is very good news and definitely a big compliment for me. Feel free to use the code. Unfortunately, I stopped working on this project due to lack of time. Therefore, I did not generate any numbers. Feel free to ask if you have questions. I will try to answer them. Best regards, Duy

…

On 19. Dec 2017, at 16:37, Andrew Hundt ***@***.***> wrote: Hey, I liked the clean way you implemented some of this code and since it is MIT licensed I was planning on merging a couple things in keras-contrib <https://github.com/farizrahman4u/keras-contrib> with attribution, but wanted to let you know! Do you have any numbers for the kind of training results you got? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#4>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHLAHy-oNEdnje23ZXamt-DOhsGfwOP4ks5tB9hUgaJpZM4RHJMt>.

[ahundt]

cool, well if you wouldn't mind reviewing my version based on yours I'd appreciate it!
keras-team/keras-contrib#197

I had to make some changes to deal with numerical issues in a way that theano would support, and I went with a simpler way of flattening the data properly while maintaining channels in the results.

I think this issue itself can be closed, thanks!