Open Azureuse opened 3 years ago
Are you sure that is not by accident? Because randomness seems play some rule in this tree search algorithm.
But, did not choosing the highest lower bound is more logical? I'm interested to test it on large MIQP problem.
For second phase of two-phase exploration the goal is to choose the leaf with the "best bound". We are performing a minimisation problem and therefore the best lower bound is the smallest one -
argmax
should beargmin
.https://github.com/oxfordcontrol/miosqp/blob/master/miosqp/workspace.py#L145
Replacing it reduces the number of iterations in my problems substantially (over 100x)!