[JVM-package] How could i use additional column just for object function calculation not for train

[zhangjinrui2718]

for example:
there is a column named pay_amount
it does not exist in feature column array but i want to use it in custom object function to calculate the precision or recall for the pay_amount

[trivialfis]

I don't think that's possible. How does a tree predict unknown feature?

[zhangjinrui2718]

I don't think that's possible. How does a tree predict unknown feature?

for example :
there is a table witch schema is
x1:double
x2:double
x3:double
c1:double
y:int
predict:int

we use x1 x2 x3 as feature columns and y as label column (binary classfication)
after training , we get predict column witch is the output of the xgboost-tree

finally we need evaluate the tree by calculating some indicators like:

sum(case when y = 1 and predict = 1 then c1 else 0 end) as sum_tp_c1
sum(case when predict = 1 then c1  else 0 end) as sum_c1

now we get the precision for c1 = sum_tp_c1/sum_c1

i find we could define custom object function and evaluate function by extendingObjectiveTrait and EvalTrait, i want to calculate precision for c1 to guide the tree splits