Guillaume Marques

Refactor solutions

1

- [x] Separate `PrimalSolution` & `DualSolution` (see #590) - [x] ~~Don't use dynamic sparse arrays anymore (ordered dict)~~ (see #609) - [x] ~~Create dedicated data structure to efficiently compare solutions~~...

update_all_ip_primal_solutions! reverse loop

4

https://github.com/atoptima/Coluna.jl/blob/50b2f4ed95dc7cfa95f4cea5cc9bddedfeefd79e/src/Algorithm/utilities/optimizationstate.jl#L223-L231 @rrsadykov what does it change if you loop in reverse or normal mode ? I don't understand the comment.

Task not completed in documentation : - [ ] more information about algorithms - [ ] anonymous constraints and variables not supported anymore

- error when creating storage in CutColGenConquer - error when executing the algorithm - preprocessing determines infeasibility but model feasible

This tolerance should be a parameter, no? _Originally posted by @rrsadykov in https://github.com/atoptima/Coluna.jl/pull/415#discussion_r527540318_

It does not look possible to do it with MathOptInterface. We have to add a constraint to the formulation to define a bound on the objective value.

Bug when `<DB= Inf>` in colgen

3

The algorithm should stop but it's not the case. ``` ```

Introduced in #721 when changing sign convention for the dual solution