Hi. I want to build a deep network composed of a couple of CNNs and an RNN. The CNNs are vggNet which can be easily implemented using Keras. However, the RNN structure is different from a simple vanilla RNN, and in the first look I found it difficult to hack into the Keras code to inherit my own custom RNN layer from the simpleRNN class. On the other hand, I think implementing this RNN using pure Theano is much easier and less complicated (based on this code <https://github.com/dennybritz/rnn-tutorial-rnnlm/blob/master/rnn_theano.py> ).
I was wondering if there is any way to jointly train the whole network when its layers are implemented using both Keras and pure Theano, and to learn their parameters together. Thanks. -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
