[R] Design Survplot performance
f.harrell at vanderbilt.edu
Thu Jul 21 20:39:56 CEST 2011
Replace the soon-to-be Design with rms.
Specify surv=TRUE to cph so that approximate rather than fully correct
standard errors will be computed by survplot/survest.
> I have a Cox PH model that's large for my server, 120K rows, ~300 factors
> with 3 levels each, so about 1000 columns. The 300 factors all pass a
> preliminary test of association with the outcome. Solving this with cph
> Design takes about 3 hours. I have created the fit with x=T, y=T to save
> model data.
> I want to validate the PH assumption by calling survplot(fit, gender=NA,
> logt=TRUE, loglog=TRUE) for many of the factors (here gender is one column
> name). Just creating this one plot takes 40m.
> I'd be happy to sample from the fitted model to create these tests, or
> figure out another way to check assumptions in the model.
> Has anyone done something similar, or have other suggestions for tests
> scale better?
> [[alternative HTML version deleted]]
> R-help at r-project.org mailing list
> PLEASE do read the posting guide
> and provide commented, minimal, self-contained, reproducible code.
Department of Biostatistics, Vanderbilt University
View this message in context: http://r.789695.n4.nabble.com/Design-Survplot-performance-tp3684580p3684626.html
Sent from the R help mailing list archive at Nabble.com.
More information about the R-help