Does it work better than original nerual network in binary classification

[study1157]

I mean apply NBDT with soft loss.., you know the inner node weight can't be update with the soft loss in binary classfication due to the special inner probability computing method.