Multiple Output for Deep Learning

[Aries Fitriawan]

Dear all,

I have just started using H2O in R programming and I really love it. Currently, I'm doing some analysis using h2o.deeplearning function.

I want to make classification model with multiple output, like y1, y2, .... yn

Can I use h2o.deeplearning function?

Cause I have done a naive test as simple as:

test.dl <- h2o.deeplearning(x = 1:875, y = 876:877, training_frame = train.hex)

x for features and y for targets (y1 and y2). But then some error messages come up :

Error in .h2o.validateModelParameters(algo, param_values, h2oRestApiVersion) : 

  Response column 'c("X876", "X877")' not found in the training frame.

I'm sure that X876 and X877 still in training frame. So how can I use h2o.deeplearning for multiple output?

Thank you :)

[arno....@gmail.com]

Hi Aries,
We don't currently support multi-target models, so you have to build separate models for each response (and don't forget to ignore the other response).
Best,
Arno

[luca.bal...@gamaya.com]

I am also interested in multi-target (multiple outputs) models for regression. I reckon it should not be quite straightforward to extend your neural network implementation to deal with these. Thank you.

[luca.bal...@gamaya.com]

EDIT: ...it should be quite straighforward....

[Arno Candel]

It’s open source, so we’d be glad to accept a pull request :)

> --
> You received this message because you are subscribed to a topic in the Google Groups "H2O Open Source Scalable Machine Learning - h2ostream" group.
> To unsubscribe from this topic, visit https://groups.google.com/d/topic/h2ostream/sjQsMV1baQ0/unsubscribe.
> To unsubscribe from this group and all its topics, send an email to h2ostream+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.