Here is what I obtained.
1- Under JMP 17 I performed all regressions and saved both Pred Formula as well as PredSE.
2- For some PredSE I adjusted to get the variance from a saved logvariance formula. This was the case for 5 responses
3- For some PredSE I adjusted to get a correct variance from a censored regression as JMP17 pro platform did not provide a predSE under censored regression. For this I first performed MLE with stepwise selection through lasso using the general regression with censor code. I then saved the regression formula and the parsimonious model. Next I performed a LS regression on the parsimonious model and saved the predSE formula to get a formula of the kind sqrt( x (XX^-1) x * var)*T(0.95,DFE). I next adjusted manually by imputing in var the MLE estimate from the censored regression. I also copied the ID from the pred formula into the predse.
4- I gather all Pred Formula and PredSE and create the profiler
5- I manually adjust the Y-scales so they represent business meaningful ranges rather than a range defined by the Y predictors.
6- I open the simulator platform, include some random uniform distributions on the factors and simulate.
results:
Misalignement between the green histograms and the profiler when the profiler is set at midpoint levels of the random uniform samplings for the factors.
7. I save a table of 10K random samples of the simulator.
8. I manually perform my own simulation by taking a copy of the original table, then go into the factor columns and substitute the original design coordinates by random uniform(low, high) according to the simulator settings. I create an additional column for each response called ' simulation formula y###' = random normal(pred formula ###, predSE ###).
9. I increase the number of rows to 5k. I now have a table that performs the simulator output manually. I compare the histogram results between my manual simulation and the 10K random samples of the simulator and they do not align. The manual simulation is according to expectation.
See attached files.
Could anyone from JMP check this?
Attached files:
20250520.SAR442970.USP.MEOH.ITER03.04_contours ::
-- mother file including all models and model functions.
-- The profiler in in script ''Profiler20250610.14H11''
JMP18 profiler simresults to table ::
-- the simulation output when performed using the profiler functionalities
20250520.SAR442970.USP.MEOH.ITER03.SIMULATION
-- the simulation output when creating the simulations by hand
Result:
Hi Yannick,
For this kind of question which is related to a strange behavior between two releases, the best is to connect with technical support. You can send them a message at support@jmp.com
Please, add me in copy so I can follow up the issue.
Thank you,
Florence
Hello Florence,
All the censoring and confidentiality is too much fuss. But if you happen to have an opportunity to inform within your company without sharing the data then feel free to be my guest.
Kind regards,
Yannick
