Inconsistent predictions with category feature

[xq-yu]

xgboost predict result is Inconsistent in a same row when a category feature was created based on different basic dataframe. For example. here is 2 different dataframe A and B, but with same columns. And assume that there is one same sample(row) both in A and B. if we tried to transform a feature from string to category(xgboost is also trained with category feature) like

A['region'] = A['region'].astype('category') 
B['region'] = B['region'].astype('category') 

the prediction may be different even if the row's value is exactly the same.

[trivialfis]

Currently, we rely on users to reuse the encoder during inference (like sklearn ordinal encoder). See our demos for an example. (demo/guide-python)

We are working on having a built-in recoder in XGBoost to memorize the encoding in the booster model #11088

[trivialfis]

Closing in favor of #11088 .