CrossValidate and Column types

[Todd Morrison]

I noticed that calling cross validate on a dataset where the target label is a REAL, there is no prediction accuracy reported. In addition, a few of the other commands also show odd behavior with this use-case.

Looking into this, I see that in GLearnerLib::CrossValidate that pLabels->relation().valueCount(0) doesn't pick up any values if the Label uses standard numerics; it only works when there is a Nominal class-style string listing.

Checking cannonical datasets (like IRIS and ZOO) shows that this is pretty standard for the ARFF test set structure.
However, when I attempt to re-target the label using the '-label' arguments, it still doesn't respond to the REAL values.

What is the expected behavior when the target label isn't a string class?
Is it just that the library doesn't support numeric labels?

Thanks!
Todd

[Anonymous]

This appears to just be a reporting issue on the CLI, since the cross validate actually does use the real values to validate.

No action required besides possibly enhacing the output in the future.