Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)Hi there!
I'm trying to plot a contour graph in JMP similar to the example below, plotted via minitab:
To make the response surface below on JMP, I saved the formula for the predicted response H in a column and plotted the following surface:
â
I would like to plot the projection of this response surface similar to the minitab example given, but I can only get the curves in the JMP surface predictor (which doesn't interest me, in this case: I want to plot it similar to the minitab example):
Plotting the contour via JMP's contour profile using the H saved prediction formula, I only get this not-at-all-satisfactory surface:
âAny ideas on how to plot this contour similar to the one in mintab? with the curves not being polygonal and broken, with gradient filling and color legend?
Thanks for the time!
â
Design of ExperimentsWindowsRe: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)I suggest that you look into the Graph Builder and it's Contour Graphing capability
The above example is using the Cowboy Hat Template sample data.
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)I plotted the predicted result of response H by the two factors of the posted surface and it was still bad:
â
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)I will put the JMP file attached to the post: https://easyupload.io/koocbm
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)If you use the Surface Profiler (available under the Profilers red triangle menu in Fit Least Squares, and under Graph > Surface Plot, you can turn on the contours
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)Yes, but this was done on the surface I posted. I wanted a 2d graph plotted as an isolated level curve, like the minitab output posted
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)Unfortunately, I do not see a way to get what you want from within the Fit Model results. Graph Builder, as indicated by @txnelson , has the ability to create the graph that you want. Unfortunately, that is not tied to a model, so you get a contour plot that is created by triangulation. This can lead to the rough appearance when you have a sparse amount of data.
So, what to do? You need a large dataset with the model so that Graph Builder will give you the smooth contours. Here is one way to do that:
From your Fit Model results, click on the Prediction Profiler red popup menu and choose Simulator (the Simulator is also available from the Contour Profiler menu).
In the dialog below the prediction profiler, change the setting from Fixed to Random > Uniform for the two factors that you want on the contour plot (check to make sure the ranges are appropriate for those factors). The other factors leave at Fixed, but put them at the setting that you wish for them to have.
Choose the number of runs for the simulation. Leaving at the default of 10,000 is probably sufficient. Under the Simulate to Table outline, click on Make Table.
You now will have a new table that you can use to create a contour plot in the fashion that you would like.
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)Wonderful, thanks for the tip! One question: why select random and uniform?
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)I would like to ask one more question: in Minitab, I have the response optimization tool to predict the best responses according to the constraints of various responses (maximize/minimize or target, for example).
I couldn't find any option in JMP for it to tell me the best results for such conditions, nor did I find where I can predict responses by typing the parameters I want in the model within the designated sample space.
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)In order for the triangulation to work and draw smooth contours, you need a grid of points. The uniform distribution will do that. Look at the resulting table and I think you will see the result.
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)You need to look at the Prediction Profiler. Optimization is under the red popup of the Profiler. From there look at Desirability. The Profiler also allows you to generate predictions for any combination of the factors. You can drag the red vertical lines to set the factors or click on the numbers in the axis area of the factors to type in the values that you want.
You really should look at the features of the Prediction Profiler. That is a fantastic tool for understanding a model.
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)A question about DSD:
Instead of inserting values to predict, I would like the opposite situation, as is done in minitab in the response optimizer. I applied DSD in the hope of generating an experiment that would screen the 16 variables to reduce them to just 3 or 4 variables. Then, I thought that from DSD I would be able to generate an RSM for the 3 responses with the same experimental runs carried out. Remembering that I would like to maximize one answer, another to minimize and another to limit to a range of values, for example.
In RSM, the objective would be to take this optimized value setting for all responses at the same time and perform the optimal welding that meets all established requirements.
(What a wonderful and active community we have here, I'm very grateful for the attention and responses!)
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)I tested here and saw that I can maximize/minimize/target for each isolated response. Would it be possible to select the maximum height considering the other two answers? For example, in order of weight and importance of the answers: smallest variation in width for the smallest width and largest height, at the same time?
In the image posted, I only see the prediction of the variables for the isolated H response, correct?
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)Yes, the Prediction Profiler allows you to do multiple response optimization. When you select Analyze > Fit Model, simply put all of your responses in the response field of the dialog box. You can then check the "Fit Separately" box so that you can create different models for each response. With this setup the Prediction Profiler will have a row for each of the responses. The Desirability Function will allow you to define the objectives for each response and then when you maximize desirability, it will be finding the best balance of all responses.
If you have already fit your models in three separate Fit Model dialog boxes, save the prediction formulas to the data table. You can then use Graph > Prediction Profiler, specify all of the prediction formulas, and then perform the optimization as outlined above.
Again, I encourage you to read the online manuals for the Prediction Profiler to see the capabilities.
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)Hi @ivanpicchi,
You can adjust the smoothness and number of levels on your contour plot with the Graph Builder by changing the cursor level next to "Smoothness" (to have curvy frontiers instead of broken lines) and adjust the number of levels (on your example by default on 4, but you can increase/decrease this number of levels).
Here are the results with raw data and predicted formula :
More infos on the details of the contour plots from Graph Builder : Contour Plots (jmp.com)
More examples on the Graph Builder capabilities can be found here : Graph Builder contour plots in JMP 15
I hiope this might help you,
Re: Help in plotting the RSM contour plot similar to the output in MINITAB (jmp 17 pro)I am also curious to know why contour lines are smooth in contour profiler but not in contour plot (under graph menu). Are both using the predicted values?
