[BUG] UnicodeDecodeError with xgboost v2.1.3 and shap v.0.43.0

[xuxu-wei]

It yields a UnicodeDecodeError when I run the the demo.

The code is literally the demo, and the versions of shap and xgboost are 0.43.0 and 2.1.3 (see Screenshots below).

The error information

---------------------------------------------------------------------------
UnicodeDecodeError                        Traceback (most recent call last)
/tmp/ipykernel_748913/3137371197.py in <module>
      9                               classification=True)
     10 
---> 11 Feature_Selector.fit(X=X, y=y, n_trials=100, sample=False,
     12                      train_or_test = 'test', normalize=True,
     13                      verbose=True)

~/anaconda3/lib/python3.9/site-packages/BorutaShap.py in fit(self, X, y, sample_weight, n_trials, random_state, sample, train_or_test, normalize, verbose, stratify)
    466                 self.Check_if_chose_train_or_test_and_train_model()
    467 
--> 468                 self.X_feature_import, self.Shadow_feature_import = self.feature_importance(normalize=normalize)
    469                 self.update_importance_history()
    470                 hits = self.calculate_hits()

~/anaconda3/lib/python3.9/site-packages/BorutaShap.py in feature_importance(self, normalize)
    713         if self.importance_measure == 'shap':
    714 
--> 715             self.explain()
    716             vals = self.shap_values
    717 

~/anaconda3/lib/python3.9/site-packages/BorutaShap.py in explain(self)
    819 
    820 
--> 821         explainer = shap.TreeExplainer(self.model, 
    822                                        feature_perturbation = "tree_path_dependent",
    823                                        approximate = True)

~/anaconda3/lib/python3.9/site-packages/shap/explainers/_tree.py in __init__(self, model, data, model_output, feature_perturbation, feature_names, approximate, **deprecated_options)
    164         self.feature_perturbation = feature_perturbation
    165         self.expected_value = None
--> 166         self.model = TreeEnsemble(model, self.data, self.data_missing, model_output)
    167         self.model_output = model_output
    168         #self.model_output = self.model.model_output # this allows the TreeEnsemble to translate model outputs types by how it loads the model

~/anaconda3/lib/python3.9/site-packages/shap/explainers/_tree.py in __init__(self, model, data, data_missing, model_output)
    973             self.model_type = "xgboost"
    974             self.original_model = model.get_booster()
--> 975             xgb_loader = XGBTreeModelLoader(self.original_model)
    976             self.trees = xgb_loader.get_trees(data=data, data_missing=data_missing)
    977             self.base_offset = xgb_loader.base_score

~/anaconda3/lib/python3.9/site-packages/shap/explainers/_tree.py in __init__(self, xgb_model)
   1685         self.read_arr('i', 29) # reserved
   1686         self.name_obj_len = self.read('Q')
-> 1687         self.name_obj = self.read_str(self.name_obj_len)
   1688         self.name_gbm_len = self.read('Q')
   1689         self.name_gbm = self.read_str(self.name_gbm_len)

~/anaconda3/lib/python3.9/site-packages/shap/explainers/_tree.py in read_str(self, size)
   1806 
   1807     def read_str(self, size):
-> 1808         val = self.buf[self.pos:self.pos+size].decode('utf-8')
   1809         self.pos += size
   1810         return val

UnicodeDecodeError: 'utf-8' codec can't decode byte 0xf9 in position 3248: invalid start byte

Screenshots