# Possible non-determinism in transcan from rms package

I ran into the following problem while using transcan() function from the rms package. For a fixed dataset, I ran transcan() to transform a set of 15 variables on 2 separate Linux machines - the final transformations were exactly inverses of each other, i.e. T1 = C-T2, where C is a constant. I am sending you plots showing the 2 transformations for one of these variables.

Would you have an idea about why transforms are differerent? Is there some non-determinism within transcan()? Are there any parameter settings that would allow transcan() to produce the exact same transformations?

My transcan() invocation is as follows:

transcan(formula_transform, data=w, transformed=T, imputed=F, pl=F)


Btw, the transcan() convergence criteria is identical on both machines.

Images of the transforms:

You sent me an e-mail about this 2-3 weeks ago which I'm sorry I haven't had time to deal with. I will do so soon. Note that transcan is in the R Hmisc package. Be sure to include self-contained code that causes the problem, in any posting, and to define versions of R and packages used. –  Frank Harrell Aug 29 '11 at 11:05