fix(deep): torch model outputs need to be transferred to cpu before c…

…hecking for additivity #3280 (#3281)
shap · Sep 27, 2023 · e51a87d · e51a87d
1 parent 1f05e2c
commit e51a87d
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/shap/explainers/_deep/deep_pytorch.py b/shap/explainers/_deep/deep_pytorch.py
@@ -215,7 +215,7 @@ def shap_values(self, X, ranked_outputs=None, output_rank_order="max", check_add
                 with torch.no_grad():
                     model_output_values = self.model(*X)
 
-            _check_additivity(self, model_output_values, output_phis)
+            _check_additivity(self, model_output_values.cpu(), output_phis)
 
         if not self.multi_output:
             return output_phis[0]