We said that there is no example code for the whole processing steps which seemed a little rash since there seem to be some gist snippets and we want to give at least credit to one:
which is minimal but at the same time very well documented.
Just a quick note which might lead to the next blog post: After we trained our network, we wanted to do a under-the-hood analysis of the reset and forget gates of the GRU cells in case of the few errors the network makes. However, due to stacking the parameters, for performance reasons, a straightforward analysis needs some more preparation. In general the question is, if we use pre-defined modules, how can we debug the internal states of individual steps and units?