Branch&bound, fixes #1 #2
Conversation
|
also, you may want to add a test with both integer and real variables. |
I wanted to add one but i wanted to use a "known" one or some that already have a solution, do you have one? |
|
Ok i added a simple MILP test which is currently failing as it did not reach the optimal value, i'll investigate tomorrow |
|
Ok pretty weird, using very large values causes unexpected results. the problem was i64::MAX causing issues, i've clamped integer values to i32 for now and it seems to work properly the problem seems to happen also when using high f64 values like f64::MAX, but it is unrelated to the integer solution I'm gonna make an issue about it |
|
I'm gonna merge this as that bug in #3 was already present so it's not a regression, i'm not sure how to tackle that for now so at least let's support integer variables for now. I added a failing test to track the bug for the future |
| continue; | ||
| } | ||
| let divergence = f64::abs(val - val.round()); | ||
| if divergence > 1e-5 && divergence > max_divergence { |
There was a problem hiding this comment.
@Specy I think you should use the EPS constant here, instead of hardcoding 1e-5.
Adds branch and bound to implement binary and integer variables, fixes #1