Profiler - Assess Variable Importance - "Variable importance requires at least two..."

Question

Hi!

Quick question, I am having trouble identifying the root cause for the message "Variable importance requires at least two rows in the data table" that appears one and then even though the data table has about 10K rows, please advise.

When using a similar dataset with mostly the same dimensions and using boosted trees, it works perfectly fine, however, if I change to another dataset, with mostly the same features, JMP keeps on returning "Variable importance requires..."

is there a specific procedure to make it work?

Many thanks!

RiztCL · Answer

Hi Victor,

You were right, I didn't share enough info, however, among the hints you gave , missing values and column contribution seems to be related to the issue, namely, if there are too many (20 out of 40) columns that were not used by the model (Boosted tree in my case) , the assessment fails , so , as workaround , if I relaunch the analysis taking those columns out, it runs ok...

I will keep on testing to give more feedback,

Thanks!

Victor_G · Answer

Hi @RiztCL,

There are too few informations to help you. Here are some questions to guide the discussions and help :

* Do you have a dataset for example, where we could reproduce the problem ?
* How was the data collected ? Random sampling, organized data from DoE, ... ?
* Did you check missing values ?
* On which platform does the profiler have a problem with Variable Importance in your case (since almost all modeling platforms have the Profiler options) ?
* Which variable importance settings did you choose : independent uniform inputs, independent resampled inputs, dependent resampled imputs, ... : Assess Variable Importance (jmp.com) ?

I think answering to these (first) questions may help Community members figure out your problem and help you with relevant advices and responses.