JMP Pro gradation

New in JMP® Pro 13

As the advanced analytics version of our software, JMP Pro contains everything users know and love about JMP – and more. With the release of JMP Pro 13, users will enjoy new capabilities, new improvements to the predictive modeling workflow and performance improvements across most of the JMP Pro platforms.

Learn how to get JMP Pro 13

JMP Pro 13 monitor montage

Text Explorer Analytics

Introduced in JMP 13, the Text Explorer is a platform for dealing with unstructured text data. While JMP 13 provides methods for basic word and phrase extraction, JMP Pro adds significant capabilities for multivariate analysis and dimension reduction, enabling users to incorporate text data into their predictive modeling tasks.

JMP Pro can take the types of text data that almost everyone has—repair logs, free-text surveys, long-form description fields and open comments, to name a few—and through the use of tailor-made analytics, convert that text data into numerical data that can be used directly by the JMP Pro modeling platforms. Now you can take the information within your unstructured data and use it to enrich your favorite JMP Pro model, be that a Bootstrap Forest, Neural Net or Generalized Regression model.

By complementing the predictors you already have with the text data, you can build models with better external validity. You are already going through the pain of storing all of this unstructured data in your database—why not put it to work for you?

Formula Depot and Generate Scoring Code

In order to compare multiple models in previous versions of JMP, prediction columns for each model had to be saved to the data table. In some cases—neural nets for example—this could significantly increase the table’s footprint. In addition, the naming convention for the new columns was tied only to the name of the column whose values were being predicted—not to the platform in which the model had been fit. This made it hard to know which platform had been used to create a given prediction formula column. Not only does the JMP Pro 13 Formula Depot address all of these needs, it also makes model deployment much easier, providing a central repository to organize, profile, compare and selectively deploy models in C, JavaScript, Python, SQL or SAS®.

Generalized Regression Improvements

The goal of JMP Pro Generalized Regression is to satisfy all of your modeling needs. Generalized Regression is the JMP modern approach to Generalized Linear Models and variable selection, as well as a world-class tool for analyzing DOEs and observational data – even life/survival data – all in one place.

JMP 13 marks Generalized Regression’s third release, and the tool continues its evolution with many additions and new features:

  • A Double Lasso option: Screens variables with an initial Adaptive Lasso pass, and uses the resulting parameter estimates as weights in a second Adaptive Lasso pass, further refining the model.
  • Addition of a Two-Stage, Forward Selection option: An initial forward selection considers only main effects, then a second pass considers interactions and higher-order terms. This technique has excellent properties for model selection in designed experiments.
  • Handles censored data, allowing variable selection when fitting survival/reliability data; support for Cox Proportional Hazards; Support for Weibull, LogNormal and Normal distributions.
  • General improvements:
    • New model selection criteria, ERIC, designed specifically for penalized regression problems.
    • ROC and lift curves.
    • Confusion matrices.
    • CDF and Quantile Profilers (like parametric survival).
    • Support for no intercept models and ordinal predictors.
    • Relaunch with active effects.Better model diagnostics.Ability to save simulation formula that can be used in the Simulate utility.

Repairable Systems Simulation

In previous versions of JMP, simulation could only be used in those reliability settings where a failure meant end-of-life for the system. However, a complex, expensive system built from many serviceable components—for example, an airplane engine—is usually repaired if possible, rather than discarded when it fails. Repairable Systems Simulation (RSS) in JMP Pro 13, analysis of repairable systems is now possible, enabling you to answer questions such as:

  • What is the mean time between failures over the “serviceable life” of the units?
  • What is the availability of the units? That is, for what proportion of time are the units functioning?
  • What are the expected repair costs over the serviceable lives of the units?
  • Which components are most responsible for system downtime and maintenance?
  • Which components should be opportunistically repaired/replaced while the system is down for another repair?

Since the RSS platform in JMP Pro 13 uses the same interface as the Reliability Block Diagram platform, reliability engineers won’t have to learn a new interface—they can get right to work building models.

General Simulation Functionality

Statisticians and analysts use simulations to evaluate new statistical methods, estimate power for nonstandard statistical tests and perform parametric bootstrapping. In JMP 13, access to general simulation capability is significantly broadened, as users no longer need to write custom JSL to perform the simulation and analyze the results. This capability is available from within most platforms having Auto Recalc or Bootstrap as red triangle menu options.

DOE is another area in which this simulation is useful. In designed experiments, the responses are not always distributed normally or approximately normally. System tests in particular can generate data that is count- or pass/fail-based. The Custom Designer in JMP 13 can now simulate realistic response data for these experiments, and JMP Pro users can use this simulated output along with the JMP Pro general simulation capabilities, to estimate the experimental power of the design.

Finally, Bagging (Bootstrap Aggregating) has been added to the Profiler in JMP Pro, making it possible for users to construct prediction intervals in settings where prediction interval formulas do not exist.Creating a validation column was largely a manual process that required multiple clicks to divide a data set into training, validation and test sets. If you required more than a simple random sample of the data for splitting, you had to use an add-in or other techniques to generate the best splits.

Now you can split data based on the problem, using one of several included algorithms. Also, if you are in an analysis platform, clicking on the validation column role without selecting a column will prompt you to choose a validation column, or create one directly from the platform. This keeps you in the analysis flow and lets you cross-validate models easier, in fewer clicks.

Hierarchical Bayes

Choice models can help companies decide which products and features are most important to customers, and how much customers are willing to pay for certain features.

A choice model that treats all customers as if they were the same will produce estimates that average the preferences across all individuals. If, as is often the case, the individuals actually have significant differences in preferences, the “optimal” product as determined by the model may satisfy this “average” customer—who does not actually exist—but possibly appeal to none of the customers that do exist.

Fortunately, JMP 13 Pro supports Hierarchical Bayes, which can properly model these differences in preferences. Use it to get better results and higher quality models from your Choice designs, maximizing the information gained from your limited run budget.

Association Analysis

Association analysis (often called Market Basket Analysis) is the identification of items that occur together in a given event, record or transaction.

Here are some examples of association rules:

  • Eighty percent of shoppers who buy product A also buy B.
  • Forty percent of repairs involving part A also involve part B.
  • Twenty percent of those having risk factors A, B, and C have condition X by age 50.

Understanding these associations can inform decision-making in a variety of contexts such as marketing, healthcare, and product reliability.

With the JMP Pro Association Analysis platform, you can also use Singular Value Decomposition (SVD), a dimension-reduction technique, to group similar transactions. The singular vectors can then be used in predictive modeling platforms.

Gaussian Process

Gaussian Process models are used to model the relationship between a continuous response and one or more predictors. In previous versions of JMP Pro, running these models was time-consuming for large data sets, and—regardless of table size—impossible for models with categorical factors. In JMP Pro 13, both of these issues have been addressed, greatly extending the reach of Gaussian process models.

Mixed Models Improvements

In JMP 13, the Mixed Model personality of Fit Model offers several new covariance structures (Unequal Variances, Exchangeable, Antedependent, Toeplitz), extending its applicability to a variety of new contexts.

Partition Improvements and Naive Bayes

JMP Pro 13 gives you more control over the Partition platform: model tuning tables let you run models over a grid of parameter values, while the randomization in Stochastic Gradient Boosting option for boosted trees helps protect against overfitting. The Naïve Bayes classifier is also available.

Documentation

Download a PDF of the new features in JMP and JMP Pro or view our online searchable documentation.

Back to Top