How to get all candidates and their scores during optimization?

Arbiter
1

i was trying to extract the Hyperparameter combination with it’s loss/score. is it possible to do so? i want to inspect all possible combination and its losses
what kind of command i can use?

i am using DL4j-example repository with specific model on
BasicHyperparameterOptimizationExample.java

https://github.com/eclipse/deeplearning4j-examples/tree/master/dl4j-examples/src/main/java/org/deeplearning4j/examples/arbiter

have been using getResult() getResultModel() on the list object of ResultReference type.

Thanks in Advance

2

Do I understand you correctly, you want to access each tested candidate, not only the best one?

3

yes, i want all candidate not only the best one.
the focus is only on the hyperparameter that is being tuned and the score.

4

You can’t get all candidates after the training is done, keeping them all can take a lot of memory.
But, if you want to collect them yourself, you can do this by implementing the StatusListener Interface.

5

Yes, you can get all candidates. Optimization uses ResultSaver interface to save all candidates generated, and you can use ResultReference interface to read the candidates after the optimization is done.
I have used these interfaces to write model structures and result scores to a database.
But if you just want to read the results in Java code, you may be fine with the file based implementation FileModelSaver and LocalFileNetResultReference that is used in the example.
See LocalFileNetResultReference and FileModelSaver

6

Thanks for the help, yes have been using the StatusListener Interface. I might take a look on the file saved based on FileModelSaver

7

You are actually right, I somehow missed that ResultSaver actually does get to save each model. Thank you very much for correcting me.