jamovi

Hi,

I'm currently developing a module focused on ML modelling. I am using the parsnip package to fit my models and I'm having trouble with the neural network function using the "brulee" inference engine which in turn uses torch as a backend. When I try to run a small model I get the "resource limit reached" error while all the other models (Random Forest, XGBOOST etc.) give no problems, my guess is that torch is too much of an overhead but I would really like to use it. Is it possible to reduce the overhead or increase the resource limit?

Thank you in advance!

Hi,

that error simply means that the R process crashed ... that's often caused by one's computer running out of resources ... in your case that's probably not the issue.

jonathon

Hi Jonathon,

I'm not sure what the issue is exactly. When I run brulee on RStudio, there are no problems, but on Jamovi even a very minimal model gives the resource limit error. All the other models (SVM, XGBOOST etc.) run without problems on Jamovi even with wide hyperparameter tuning grids. Additionally brulee requires to install torch manually which has to be done on Jamovi source R files, maybe there is another way to install torch binaries to reduce the overhead. If there is no alternative I can switch to the nnet engine which gives no problems but I would really like to use brulee/torch.

Thank you in advance,

Javier