Now has the option when creating a TrainableNetwork to enable sending summary data to TensorBoard. This became rather complicated since the actual saving of information is not a graph operation like all the other TensorFlow things we are calling from Java. This means that Java has to handle creating the file. The summary merging is a graph operation. Once this is fetched from Java, the information is extracted and reformed into an Event that is serialized using protobufs compiled from those in the TensorFlow directories.
Note that to use TensorBoard, you still need to run tensorboard --logdir ./logs/.
Now has the option when creating a
TrainableNetwork
to enable sending summary data to TensorBoard. This became rather complicated since the actual saving of information is not a graph operation like all the other TensorFlow things we are calling from Java. This means that Java has to handle creating the file. The summary merging is a graph operation. Once this is fetched from Java, the information is extracted and reformed into anEvent
that is serialized using protobufs compiled from those in the TensorFlow directories.Note that to use TensorBoard, you still need to run
tensorboard --logdir ./logs/
.