JMP® Pro

The Neural platform includes techniques to help you mine and model data with greater accuracy and flexibility. You decide which data to use for cross-validation. You get rich diagnostics and the ability to fit one- and two-layer networks with your choice of activation functions. You can also automate the handling of missing values. Bootstrap forests and boosted trees in the Partition platform allow you to safeguard against making predictions that are overly optimistic or skewed by the specific data used. This means the predictive models you build are more likely to generalize well, often with improved classification accuracy as quantified by the usual measures of association.

Learn more about JMP for Interactive Data Mining

Exact tests enable you to make reliable inferences when your data is small, sparse, heavily tied or unbalanced, and the assumptions and approximations of the corresponding large sample theory are in doubt. For contingency analysis, JMP Pro provides exact versions of Fisher’s test (for association between rows and columns), the Cochran-Armitage test (for trends in binomial proportions across the levels of a second variable) and the Kappa statistic (for testing agreement between two variables that have the same levels).

In the Oneway platform, JMP Pro provides exact versions of the Wilcoxon (Mann-Whitney), Median, van der Waerden and Kolmogorov-Smirnov tests whenever the X variable has two levels. All exact tests are integrated in the Fit Y by X platform, and they become available only when the selected Xs and Ys satisfy the correct conditions. This allows non-expert users to utilize inferential methods that will remain reliable in any real-world situation.

Splitting your data into chunks is the best way to assess how well the predictions from the models you build will generalize. Each chunk is used to support a different phase of the model-building process (training, validation and possibly test, too). JMP Pro extends this approach to stepwise regression in the Fit Model platform, introducing a Max Validation RSquare stopping rule for selecting appropriate model terms either automatically or manually. In addition to enriching your model-building approach, splitting your data consistently also enables you to more easily compare and contrast competing models, from the same or different families. To further facilitate such comparisons, JMP Pro also provides the Confusion Matrix when performing nominal or ordinal regression using Fit Model.