Reputation: 279
I am building an RNN (LSTM/GRU) and I have time series of variable lengths. How can I reset the state variable after each time series is done? I would like to do this in Keras, but if I have to build an RNN in Theano (my back-end) that's fine too.
Upvotes: 0
Views: 1095
Reputation: 21
In Keras Documentation, Chapter on Recurrent Layers:
To reset the states of your model, call .reset_states() on either a specific layer, or on your entire model.
E.g.:
model.predict()
model.reset_states()
Upvotes: 2
Reputation: 279
It isn't in Keras' documentation but recurrent layers have a .reset_state() function. You can find it on the github and you can either reset the states of one layer or the entire model.
Upvotes: 3