Just got off a call with a customer (Nvidia) and this idea came up during the call. The customer asked how to use JMP to identify tests with similar test results (pass/fail). The datatable is many tests as the columns, and different GPU chips being tested as rows with a 0/1 indicator of pass/fail for a given test.
Showed the customer the Cluster Variables platform which got them very excited as it did exactly what they wanted, ie. identify tests with similar results so that they can remove redundant tests and reduce test time without impacting quality. There were however a couple a key features which would have made this platform much more usable for this type of application.
1. The platform needs to allow you to ungroup a member from a cluster and re-compute the "Proportion of variation explained by clustering" metric. Example you should be able to right-click on the Cluster Members report to do this. You would want to do this for columns which have weak correlations within their assigned cluster. For the application above, this would mean keeping tests with weak correlations in their cluster in the semiconductor chip test flow.
2. You should be able to define a target value for the "Proportion of variation explained by clustering" metric and JMP would calculate the minimum number of clusters necessary to achieve this, and give you a plot of POVEBC versus cluster count. The would help the test engineer identify how many tests are necessary to capture a certain percentage of all the failures.
