Applying the XGBoost model to new dataI followed your Powder Metallurgy example but used my own data (the first of 10k rows of a larger data set). I built a model with objective="binary:logisitic" and saved the prediction formulas to the data table. Then I attempted to make new predictions by applying the model to 10k new records which did not contain the "Y, Response" variable by concatenating these new rows to the bottom of the first 10k rows, but the formulas weren't able calculate the "FoldA==1" etc columns for the new rows nor give my a prediction probability. Do I have to use a continuous validation column, or is the issue related to something else?
Thanks
Re: Applying the XGBoost model to new dataYes, should work fine, and we have a fix implemented in JMP Pro 16 Early Adopter that should make the 0 setting unnecessary. Another term for this is "repeated k-fold", and it is one of the very best ways to rigorously assess predictive models.
Re: Applying the XGBoost model to new dataHi @russ_wolfinger,
If you're doing multiple-fold validation, e.g. using all three folds, would you fill to end of table FoldB and FoldC as 0 as well?
It appears this is what's needed, but just want to confirm.
Thanks,
DS
Re: Applying the XGBoost model to new dataHi Nate, For now please fill in FoldA with the value 0 for all of the new rows and the formulas should then propagate through. One way to do this in your existing table is to double click the cell from FoldA for the first row of new data and change it to 0. Then Right click > Fill > Fill to end of table. For the next version we'll plan to make sure the formulas handle missing values for the fold variables.
