Motivation

The original implementation takes best_params as the x fetched by argmax(Y[is_feasible]).
However, what we really want is x with the best feasible y.

See the example below,

X = [[1.0], [4.0], [2.0]]
Y = [2.0, 3.0, 5.0]
C = [1.0, -1.0, -1.0]

In this case, feasible = [False, True, True], meaning that Y_feasible = [3.0, 5.0] and X_feasible = [[4.0], [2.0]].
So X[argmax(Y_feasible)] becomes [4.0] instead of [2.0], which is the actual best feasible params.
I fixed this bug in this PR.

Description of the changes

• See the motivation