LOOCV in Datarobot

[MManojkumar]

Hi Team, I know we have option for exercising Cross validation. I have a dataset which has only 90 rows, which i cant increase for sure. I am planning to do Leave one out Cross validation (LOOCV), i tried searching in datarobot forums. Can anyone please help me how to do that?

[Erica · Accepted solution]

Hi Manoj

As Far as I am aware, in my version of DataRobot, the only options are for K Fold Cross validation, with an option to adjust the number of folds from 2 to 50, or TVC.

Leave-out-one-cross-validation is basically an extreme form of K-Fold validation, where it is repeated for every point in the data set - IE, K=N (number of points)

The closest you will get to that here is to increase the number of folds in data robot to 50 which isn't far off 90!

However

I suspect there's a good reason that LO1XV hasn't been  included in the DataRobot software - a quick search yields warnings about high variance amongst your error estimates. Not to mention  it's probably computationally expensive for most data sets that we would be likely to use.

Hope this helps!

Erica