Ability to set loss function by column


Madeleine suggested giving the user the ability to set the loss function GLRM uses for each column of the training frame. One numeric column may have a quadratic penalty, while another be better fit with a Poisson distribution, so it is not sufficient to assume all numeric (or categorical) columns are treated in the same way.

This requires some thought on the API. It would be similar to selecting included/excluded columns during training, except with the additional complication that the user can choose which loss function to use for a column from a drop-down menu. An additional tool/button should be included to allow the user to set all numeric columns to a specific loss function, e.g. quadratic.


Former user
August 18, 2015, 8:15 PM

Blocked by PUBDEV-1914. Multinode tests fail because H2O cannot serialize enum arrays.

Your pinned fields
Click on the next to a field label to start pinning.