Chemists, and Pharma drug formulators really like to use mixture designs, and they love that they can combine Mixture and Categorical factors in the Custom Design platform. Not being able to generate compelling visualizations of their data is frustrating.
The customer's solution was to simulate lots of points and apply color gradients to the data points so they could see a gradient fill pattern. However, since the number of points necessary to to get a clean fill was very high, JMP became very slow and laggy. The solution was to switch to Spot Fire to visualize data from JMP. (very bad) People in this sector don't complain, they find other solutions and move on.
Ternary plot wish: Rebuild ternary plot so that it is in graph builder, or is graph builder like and has all the controls like graph builder.
Problem 1: Filled contours rely on the Contour Function, but the option to fill the contours is not exposed. With some trickery, a user can force the contour function to be included in the graph script, and then the filled contour, "Filled(0)" can be discovered in the script.
Problem 2: The Contour Function only works with continuous variables. So if a customer uses Custom DOE to make a design that includes both categorical and mixture factors, then saves the prediction formula to use in as a Contour Formula in a ternary plot, then tries to make a ternary plot, it appears broken, no contour lines only data points. There should be sliders or selection boxes for the categorical variables, and of course there should be contours.
Problem 3: The fill colors for constraints are essential three different shapes, one from each side. When the transparency of the fill is decreased the "corners" of the triangle are very dark because there are overlapping shapes. If we want to see what the contours look like outside of the constraints but also show the constraints, the figure looks very odd. The constraints should fill the ternary plot like a triangular contour rather than a shape extending from the axis.
I'm sure this is a difficult legacy plot so I appreciate that any changes might also be difficult.
This is a horrible, but functional solution to the filled contour with categorical variables in the DOE problem:
1. Rename the levels of your categorical factors to be numbers. Level 1 is 0, level 2 is 1, level 3 is 2 etc. (I know, makes the model interpretation much harder, but…)
2. Change the data type to Numeric and modeling type to nominal.
3. Run the model, save the prediction formula to column.
4. Change the categorical variable columns to continuous. (I know super wrong, but…)
5. Graph>Ternary plot: use the prediction col for the contour formula, and your mixture variables as the X’s
6. Now you get sliders for the categorical levels that are pseudo coded as continuous variables, and you get contour lines.
But wait there’s more. In the next step we are going to force the display of a contour function. Look up “contour function” in the scripting index.
7. Change one of the contour values. Then save the script to the script window.
8. In the script window you should see the JSL for the ternary plot, with the entire contour function, as well as some other arguments.
9. The second to the last line of the contour formula is “Filled(0)”. Change that “0” to “1”.
10. Run the script.
