Closed t13m closed 8 years ago
Hi @t13m, tkx for the detailed report.
I think I pushed something to Keras doing that already: https://github.com/fchollet/keras/blob/master/keras/layers/recurrent.py#L76 Did you try updating Keras?
Let me know how ConvGRUs worked out for you.
Hi @EderSantana , the test passed after updating Keras. Thank you.
Hi, I ran the unit tests but failed on convGRU. I found that
Recurrent.get_output()
will be executed when compiling the model, which contains followed lines (Line 69 and 70):The
self.input_dim
here isNone
and theself.output_dim
is(1, 28, 28)
. So thereducer
can not be created correctly.I'm not sure if I got something wrong. But I copy-pasted the
get_output
method toconv_rnn.py
, and modified those lines like this:And then the code can pass the tests.