You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Using Blended Conditional Gradient (BCP) as node solver produces non integral vertices for some instances of the sparse regression, portfolio and lasso problem.
The issue seems to stem from the separation oracle.
The text was updated successfully, but these errors were encountered:
The issue is not with the separation oracle but with the simplex_gradient_descent_over_convex_hull. Here, we optimise over the current active set. Returned in the Frank-Wolfe callback is a point that is itself a convex combination of the active set. Thus, nearly never integer feasible.
This is expected and we deal with this in Boscia by ignoring "vertices" from BCG there state.tt=FrankWolfe.simplex_descent.
Using
Blended Conditional Gradient (BCP)
as node solver produces non integral vertices for some instances of the sparse regression, portfolio and lasso problem.The issue seems to stem from the separation oracle.
The text was updated successfully, but these errors were encountered: