Deeplearning4j on spark: Is it possible to capture weights for each layer while training neural network

Question

I have been trying to find if there is a way to capture and save(as json) weights of each layer of neural network while training using DL4J(deeplearning4j on spark).

Please let me know if someone has any idea.

Answer 1

During a training loop you get back a multilayernetwork from fit: https://github.com/deeplearning4j/deeplearning4j/blob/master/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/impl/multilayer/SparkDl4jMultiLayer.java#L202

That's a local version of the model youc an do whatever you like with. See http://deeplearning4j.org/spark for more.