Issue Encountered When Using DeepHit for Survival Prediction

[juinzhu]

When applying DeepHit for survival prediction, I encountered the following issue:

lrns[[i]]
LearnerSurvDeephit:surv.deephit: Neural Network

• Model: -
• Parameters: list()
• Packages: mlr3, mlr3proba, mlr3extralearners, survivalmodels, distr6,
  reticulate
• Predict Types: [crank], distr
• Feature Types: integer, numeric
• Properties: -

lrns[[i]]$train(lungtask)
Error in py_module_import(module, convert = convert) :
ImportError: dlopen(/Users/juinzhu/miniconda3/envs/p39/lib/python3.9/site-packages/torch/_C.cpython-39-darwin.so, 0x0002): Symbol not found: _ZN3c1010ReplaceAllERNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEENS0_17basic_string_viewIcS3_EES9
Referenced from: <552ADBFF-CB85-30D2-9765-6C0C3EB625D7> /Users/juinzhu/miniconda3/envs/p39/lib/python3.9/site-packages/torch/lib/libtorch_python.dylib
Expected in: /Library/Frameworks/R.framework/Versions/4.4-arm64/Resources/library/torch/lib/libc10.dylib
My Python version is 3.9.7, with PyTorch 2.6.0.

The R version I am using is 4.4.2, and in R, the version of PyTorch installed is 0.14.2.

What should I do?

[cregouby]

Hello @juinzhu

There is a confusion here between {torch} that has nothing related to python nor pytorch, and your issue.

As your log mention miniconda, py_module_import() and libtorch_python, I would recommand you to open your issue either in {survivalmodels} or one of the other of your package.

As a general advice, publishing a reproductible example with reprex::reprex() would have shown directly that you don't rely on {torch} in your issue.

Hope you will find a solution to your issue.