Missing data

[micha-hro]

I fitted a XGBoost model using data that contains missing values.
For each feature, the tree dump does define to which group cases with missing data belong:
0:[f2<3] yes=1,no=2,missing=1,gain=138.535736,cover=33.6236076
However, the supertree visualization handles cases with missing data in such a way that they are counted in both leafs.
Is this intended?

[pplonski]

Hi @micha-hro,

I think missing values should be displayed in the edge with value 1 , the same as yes values. If it is otherwise, it might be a bug.

You can check alternative ways of decision tree visualization from Xgboost, you have them described in this article https://mljar.com/blog/visualize-xgboost-tree/

[micha-hro]

I came across the supertree visualization through this link. :)
It seems to be a bug, because the missing cases are counted for both yes and no.