Subscribe Bookmark RSS Feed

modelling number of occurence to two input values

Hello all!

I am trying to model the number of error occurrence to two input values. My table contains:
pressure: analog
radius: analog
error: ordinal (0/1)
and some index for type, date, run of experiment.

There is many lines with same pressure/radius combinations that may have an error value of 0 or 1. The table contains roughly 200.000 entries.

Which functionality can I use for getting an answer to which pressure/radius delivers the least errors? Basically I would want to plot a response surface in the end.
One of my problems seems to be that I simply do not know the correct wording in english (I am native german speaker). also my jmp runs in german and I cannot find a switch to english. Interestingly my help file is in english!?

Can someone please give me a pointer?

Thanks a lot!
3 REPLIES
laural

Community Trekker

Joined:

Jun 23, 2011

I think the first thing to do a table summary on the pressure, radius and sum up the errors.

Next create a new column that is the sum(errors) divided by the Nrows (assuming there are no missing values for errors). Now you have a percent column that you can plot as the response for your two factors.
ms

Super User

Joined:

Jun 23, 2011

You can try a logistic model using the Fit Model Platform. Before you start, change error column to nominal. Set error as Y, pressure and radius (and the combination pressure*radius if relevant) as effects and run the model.

Then you can use the prediction profiler to observe what levels that minimize the error probability. There are even some built-in optimization algorithms that can be invoked under the profiler red triangle (I have not tried them though).
Thanks for the pointers, I did both.

summary
The summary is easy going. It basically yields an error rate percentage. I plotted those as contour diagram and got what I was looking for. The response surface diagram works as well. The alternative I see there is the Pareto Plot (pressure and radius are properties of discrete exchangeable machine components). They all yield matching results.
Percentages are around 0.12% which is quite high for my application.

logistic model
input is the original data table, error is nominal.
There I see the possibility to get information to which pressure/radius combination could be optimal even if it has not been among the test candidates.
Here I got one question: How do you enter pressure*radius into the model dialog? All I see is single column entries!? How do I enter that?
The result I got using only radius and pressure by themselves yields two graphs with vertical lines at the pressure and radius levels where I have test candidates. There are 3 and 5 levels of pressure and radius respectively. There I must admit I fail to interpret the graph. What does it mean to have those points in vertical lines?

Thanks a lot for the input!
-Jean-