Sparsity enhancement not working

[mushroompasta]

Hi,
I'm running DicE with the Pima Indian dataset (8 predictors, all continuous). I've been trying out both the genetic approach with sklearn and the gradient-based on with TF2, but the problem with both is that the returned counterfactuals are not sparse. I've experimented with the params proximity_weight, diversity_weight and sparsity_weight, but even setting both proximity_weight and diversity_weight to zero and sparsity_weight to some large value doesn't have an effect -- the resulting counterfactuals always change almost all of the features. When I specify features_to_vary to e.g. three features, then I do get counterfactuals with only those features changed, so in that case it works. But I don't want to specify the features to change, I just want the results to be sparse.

Does anyone have a clue what could be causing this?