I would like to report a potential bug that appeared when I ran a model using the tensor smooth t2() on different systems:
I use the model
model <- brm(Response ~ 1 + t2(A, B), data = mydata, family = student)
where Response, A, and B are z-scaled continuous variables.
I fitted the model on three different systems:
Sys1: Windows 10, platform x86_64-w64-mingw32, R version 4.0.0 (2020-04-24), Brms version 2.12.0, local machine
Sys2: Linux, x86_64-pc-linux-gnu (64-bit), R version 3.6.3 (2020-02-29), Brms version 2.12.0, virtual machine
Sys3: Linux, x86_64-pc-linux-gnu (64-bit), R version 3.6.3 (2020-02-29), Brms version 2.12.0, server cluster
On Sys3, the raster plot output of
conditional_smooths(model) looks like this
but on Sys1 and Sys2 the output looks like this:
As you can see in the above example, the y-axis is mirror-inverted. The latter example turned out to be the correct one.
I’ve made sure that the axes are not inverted and that the data are the same – I even saved the model output from Sys3 and ran it on Sys1 and Sys2 using this model:
replication <- brm(model$formula, model$data, family = student)
to make sure that the data were not accidentally inverted.
So far I can’t rule out a bug only affecting Sys3, yet the only difference between Sys3 and Sys2 is that Sys3 is a large server cluster with a high number of other R-packages installed. Both Sys1 and Sys2 are my personal machines.
Please let me know if there is anything I can do to help getting to the bottom of this issue.