Replies: 1 comment 2 replies
-
Small update: not only in terms of computational performance, but also in terms on functional behaviour, ntime does not seem to make a difference when passed as a vector of discrete values. It looks like it is parsed correctly in the R code but it seems that the trees are grown on all times (I get the exact same splits and the exact same predictions no matter how I set ntime). I think I can get the behaviour I want (and a decent computational time) by manually manipulating the eventtime (ceiling) and manually applying administrative censoring in the data), but I would appreciate a clarification on the expected behaviour of ntime. Thanks!
|
Beta Was this translation helpful? Give feedback.
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
To speed up calculations, and as I am not interested in predictions at all event times in the dataset, I specify ntime for both survival and competing risks models. (ntime = 1:7)
I would expect that the logrank split statistic is calculated by summing over these 7 ntimes (or closeby proxys from the provided eventimes).
I would expect it to reduce the runtime as in the situation when I provide discrete eventimes in the dataset (1,2,3,4,5,6,7).
There is though a big difference between the two situations. The difference is larger for competing risks.
As I am working with large datasets and running a lot of models, working only with ntime (without rounding the eventtimes with ceiling) is not feasible at all for me.
Of course, I might have wrong expectations about what ntime does; my expectation that it would run as fast as in the situation of discrete times might be wrong.
I would appreciate an explanation.
See below an example with the runtime on my Windows machine in comments.
Thank you!
Beta Was this translation helpful? Give feedback.
All reactions