Hello,
I was wondering if there are any materials in regards to monotonicity constraints in the multinomial case? If so, how does it work with multiple class outcomes?
Note, I did find this issue: https://github.com/dmlc/xgboost/issues/3248 but wondering if there was a follow up?
Cheers,
Navdeep
Can you elaborate more on what you mean by multinomial monotonicity constraints? Right now each individual tree is constrained to produce monotonicity of a score output with respect to a chosen set of features.
Sure. I mean if I run an XGBoost model to predict a multinomial outcome. For example, the iris dataset. Now, how does the monotonicity constraints play out for this particular problem?
Hello, according to this user
Monotonic Constraints are not available for multiclass classification problem. Actually they give both different results wether they are used or not, plus XGBClassifier and xgb.train give different results (I assume XGBClassifier does not support monotonic constrains)
Here I expose the problem
https://stackoverflow.com/questions/59228014/xgbclassifier-wrapper-v-s-xgb-train.
If I include constrains, results differ (not applied to XGBClassifier wrapper)
