no attention mechanism, what's the difference?

[mw66]

Hi,

(This is more a discussion / question than a bug.)

Looks like this implementation doesn't have the attention mechanism (align to the ascii char), but still able to generate writing. I wonder what's the difference it will make if it has?

e.g. https://github.com/greydanus/scribe

[hardmaru]

Try to add the conditional generation bit, and let us know :-)

…

On Mon, Apr 3, 2017 at 10:19 PM mingwugmail ***@***.***> wrote: Hi, (This is more a discussion / question than a bug.) Looks like this implementation doesn't have the attention mechanism (align to the ascii char), but still able to generate writing. I wonder what's the difference it will make if it has? e.g. https://github.com/greydanus/scribe — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#20>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGBoHvlf6ai48ZOE4Zj6t7LYZIFhXKuJks5rsdLggaJpZM4MyZZp> .

[mw66]

But need to add the attention mechanism (one-hot for each char in the seq) to the training first.