Multiple outputs

[eker...@gmail.com]

Algorithms like neural networks can obviously have multiple outputs, can I specify multiple outputs in H2O (e.g. for the deep multilayer perceptron)?

Thank you.

[cli...@0xdata.com]

Yes; you can train on a dataset with multiple output classes; we've seen Deep Learning do well even with 100 output classes.
Cliff

[eker...@gmail.com]

In the web UI there's a select field for "response" and I can choose only 1 option... what am I doing wrong?

[eker...@gmail.com]

What I'm trying to accomplish is regression with multiple outputs (specified as multiple columns), but I can't find a way to add multiple "responses". It's not multiclass classification.

[Arno Candel]

Hi,
I see. H2O doesn’t currently support multiple response columns. There was some initial work on this, but the added complexity to the scoring logic would make this somewhat hard to integrate smoothly. You are welcome to make changes to the code, if you’d like, we can point you to the right place(s).
Arno

[cli...@0xdata.com]

Right now the only way to do this is to train a new model for each response column.
Cliff

[marco....@gmail.com]

Hi,

in the web Frontend it is still only possible to have one response. Are there any ways to train multiple response columns? Or am I getting something wrong in the prediction of a multidimensional feature vector?

From some known parameters I want to guess the most probable scenarios of the missing parameters.

[Erin LeDell]

You can have multiple response columns in your data frame, but you can
only use one of those columns as a response when you train a model.
This makes it easy to train models with different responses -- you just
have a separate model for each one.

To deselect the additional columns from being used in the training set
in Flow, just click on them in the "Ignore columns" section once you
start the modeling process.

-Erin

--
Erin LeDell Ph.D.
Statistician & Machine Learning Scientist | H2O.ai

[marco....@gmail.com]

Oh, thanks for that very quick answer! As far as I understand Geoffrey Hinton in his nice videos, Deep Belief Networks should also be working with multiple response - anyway H2O is cool stuff, I wish you the very best!

Greetings
Marco

[Aries Fitriawan]

Hello

Currently I'm doing something with h2o in R programming

And also I want to have a multiple output, like y1, y2, .... yn

then I tried something like this

test.dl <- h2o.deeplearning(x = 1:95, y = 96:100, training_frame = train)

then come up with this error message

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

  Response column 'c("X96", "X97", "X98", "X99", "X100")' not found in the training frame.

Anyone know what does it means?

thank you

[Spencer Aiello]

H2O algorithms currently support at most one target column. For multiple labels, you'll have to build individual models for target column.