This is a question for keras, not Theano. The problem code from change in Keras, not Theano so we can't help.
Keras have a tendency to break compatibility. I don't know if there is a fix except using the old keras, write to them, maybe they can help you. Fred On Tue, Jan 31, 2017 at 9:03 AM, Kezhi Li <[email protected]> wrote: > Previously I used keras==0.1.3, and trained lots of models (LSTM model > with files format '.h5' in my case). Let's say they are 'OM' (old models) > trained from 'OK' (old keras). Every time, I generated the model structure, > and used 'model.load_weights' to load the weights. All work perfectly. > > Now I upgrade to keras==1.x.x (eg.1.2.1), and try to use the new version > of keras to do the same as before. I generate the same NN (changed codes a > little), and run the same LSTM sequence prediction codes as before. > 'model.load_weights' can be run smoothly, however, all predictions are > totally wrong. > > Then I use the keras==1.x.x to train new model and save its weights. Let's > say they are 'NM' (new model) trained from 'NK' (new keras) > > So far, what I found is > > 'OK' can load 'OM', the prediction/output is correct > > 'OK' cannot load the 'NM' weights, the error is: KeyError: "Can't open > attribute (Can't locate attribute: 'nb_layers')" > > 'NK' can load the 'OM' weights, however, its prediction/output is totally > wrong > > Because I have trained lots of models using 'OK', I really hope the models > are not useless. But if I stick to the 'OK', some of functions or methods > of 'NK' are not available, in addition all menus are written for 'NK'. > > Is there any one know what happened here? > > 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. > -- --- 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.
