Dzmitry Bahdanau
Dzmitry Bahdanau
I think most people prefer to print status, because without printing it one doesn't even know how many iterations of training have been done.
+1 On 14 December 2015 at 01:42, Tom White [email protected] wrote: > Any objection to having the latest official release listed on the Python > Package Index (PyPI)? I think...
Yep, I am going to inherit from one of those.
Yep, I also do not understand why fan-in is not a part of the parameter's shape. On 6 April 2016 at 05:38, Sigurd Spieckermann [email protected] wrote: > Can you elaborate...
Actually you are right, since we even don't have an expression that works for fully-connected and convolutional layers, let alone all possible crazy layers of the future. We can have...
I don't understand, now you have removed your fix, and it is again not supported to have no states property. Why not just implemented like you did in the first...
Are you running the test on GPU? I assume so, because Travis seems to be quite happy with your PR. Can you please share the error message with us, so...
Seem like you right: that should be `BoundApplication`. If `recurrent.py` was better covered by tests, that would emerge as a bug. I don't think we are missing anything major. For...
Permuting a few sections in the documentation seems to be a CCW level task.
If this is really helpful, we should make it mandatory, I think. Just as currently all the initial states of recurrent networks are trainable, which is not a completely standard...