Raised error for one-hot encoded data with generic method

[Gradwanderer]

I have preprocessed a data set with one-hot encoding for all categorical features. The data has now more than 30 binary (hot encoded) (0,1) features. If I try to run the "genetic", I get the error shown below.

Do I need to specify the (0,1) encoded features?
The change to random does provide CFE, but I guess the "random method" is not the optimal solution to get the best possible CFE.

File "C:\CFE.py", line 158, in gen_cfe
e1 = exp.generate_counterfactuals(downsampled_dataset, total_CFs=10, desired_class="opposite")# "opposite")
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:.venv\Lib\site-packages\dice_ml\explainer_interfaces\explainer_base.py", line 186, in generate_counterfactuals
res = self.generate_counterfactuals(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:.venv\Lib\site-packages\dice_ml\explainer_interfaces\dice_genetic.py", line 270, in generate_counterfactuals
self.num_output_nodes = self.model.get_num_output_nodes2(query_instance)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:.venv\Lib\site-packages\dice_ml\model_interfaces\base_model.py", line 70, in get_num_output_nodes2
return self.get_output(input_instance).shape[1]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:.venv\Lib\site-packages\dice_ml\model_interfaces\base_model.py", line 54, in get_output
return self.model.predict_proba(input_instance)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:.venv\Lib\site-packages\pyod\models\base.py", line 213, in predict_proba
test_scores = self.decision_function(X)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:.venv\Lib\site-packages\pyod\models\pca.py", line 300, in decision_function
cdist(X, self.selected_components) / self.selected_w_components,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:.venv\Lib\site-packages\scipy\spatial\distance.py", line 3006, in cdist
return cdist_fn(XA, XB, out=out, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: Unsupported dtype object

[Gradwanderer]

Just to add some "problems" to my setting.
If switched to random sampling, the Dice package does generate CFEs that are not possible.
For example I have the feature sex with (male, female, other) encoded in Male (0,1) and Female (0,1). This solution would require a datapoint with [0,0] for a "other" choice. A datapoint with [1,1] does not exist. I get such solutions, because I do not know how to specify how the change can be done. Is there a solution to that specific problem or do I need block a cange in such features?