Statistical FAQs
How are the crossvalidation statistics defined in Partition?
The k-fold crossvalidation method randomly divides all of the (nonexcluded) rows in the datatable, D, into k subsets: D1, D2, . . . , Dk. Each row is randomly assigned to one of the k groups.
Note: This means that for k=n (the sample size), some groups will have several rows, and many groups will have none.
K different models are then trained using data from D – Di, where Di is the holdout fold. For continuous responses, the error for each observation in Di is calculated using the model, trained from D – Di. For nominal and ordinal responses, JMP calculates -2LogLikelihood for each observation in Di using the model, trained from D-Di. This is repeated for each of the k folds. The resulting errors are squared and summed (or in the case of nominal and ordinal responses, the -2loglikelihood values are summed) to construct the crossvalidation SSE (or crossvalidation -2loglikelihood in the case of nominal and ordinal responses).
FAQ # 2097
Last Updated: 2005 Jun 30
Browse JMP FAQs
Support Documents
Contact JMP Support
919.677.8008 (US)
