NorskRegnesentral / shapr

Explaining the output of machine learning models with more accurately estimated Shapley values
https://norskregnesentral.github.io/shapr/
Other
147 stars 34 forks source link

VAEAC with old Cuda #392

Open AbdollahiAz opened 7 months ago

AbdollahiAz commented 7 months ago

Hi shapr,

I prepared a code with following vaeac configuration and ran it on an HPC. Unfortunately, our HPC is currently support Cuda no more than 11.3 version. I added this line to my code vaeac.extra_parameters = list(vaeac.cuda = FALSE) but I faced the following error.

 expl_vaeac_without <- explain(
   model = model,
   x_explain = x_explain,
   x_train = x_train,
   approach = "vaeac",
   prediction_zero = prediction_zero,
   n_batches = 1,
   n_samples = 250,
   vaeac.epochs = 50,
   vaeac.n_vaeacs_initialize = 4,
   vaeac.extra_parameters = list(vaeac.cuda = FALSE) 
 )
Note: Feature classes extracted from the model contains NA.
Assuming feature classes from the data are correct.

Error in setup_approach.vaeac(internal, model = model, predict_model = predict_model) : 
  `torch` is not properly installed. Please run `torch::install_torch()`.
Calls: explain ... setup_computation -> setup_approach -> setup_approach.vaeac
Execution halted

Please let me know is it possible run the vaeacapproach on this HPC?

Thanks in advance, Az