Neural network model prediction in Keras without using GPU -
my neural networks model built in keras on theano using gpu. storing using pickle future use, possibly on computer.
is possible use model prediction without gpu?
sure. it's common use-case. gpus boost training, aren't available in production (for example, if run on customer's phone).
i don't know theano much, might have equivalent tensorflow.serving
. can serialize trained model object , read other machine.
to serialize, can either use:
- the built-in
keras.models.save_model
,keras.models.load_model
dumps models hdf5 files. - if need/prefer pickle - it's not supported keras, can use trick - http://zachmoshe.com/2017/04/03/pickling-keras-models.html
Comments
Post a Comment