Rstudio on EC2 instance crashes when torch is called about tabnet HOT 8 CLOSED

I think this can be related to a problem in the torch installation. Can you check that you can do torch::torch_tensor(1) in that machine?

from tabnet.

Hi,

Its giving the same error msg when i do that. "The previous R session was abnormally terminated due to an unexpected crash.

from tabnet.

Hello @G3Jram ,
so here you encounter a torch:: installation issue, not a tabnet:: issue. Torch is the underlying deeplearning framework of Tabnet. I think it is worth for you to reoppen the issue in here: https://github.com/mlverse/torch/issues, if the problem remains after you check all the installation prerequisite in https://torch.mlverse.org/docs/articles/installation.html.
Hope it will helps you enjoy using Tabnet.

from tabnet.

Hi,

I have succeeded in implementing tabnet in python instead. Waiting for the torch error to get sorted on the R side..

However, after building the model if i want to look at the feature importances, im not too sure how to do it.

Tabnet doesnt need shap to understand the features right?..

I actually want to plot the features by importance.

Please help.

from tabnet.

Hello @G3Jram ,
the R implementation includes the tabnet_explain() function with associated plot that will give you feature importance for each prediction as documented in https://mlverse.github.io/tabnet/articles/interpretation.html. This provides you the internal model feature importance, so you don't have to suffer the drawbacks of surrogate models like shap.

from tabnet.

we have a post on that in fact: https://blogs.rstudio.com/ai/posts/2021-02-11-tabnet/

from tabnet.

Hi,

Sorry if i wasnt clear. I am after the feature importance in python not R for now as i still face torch installation issues in R.

Any help in python would be good.

Thanks.

from tabnet.

You can find examples for the python API here: https://github.com/dreamquark-ai/tabnet/blob/develop/forest_example.ipynb

Also, I think for questions on the Python interface It's better if you ask in their repository: https://github.com/dreamquark-ai/tabnet

from tabnet.