bfpc
commented
1 year ago @bfpc I'm reverting #531 because it doesn't account for cases with integer variables (so no dual solution will be present).
The place to fix it is probably here:
I had thought about that, but I thought it might result in inconsistencies, since the primal solution found might not correspond to the dual solution if the latter gets recalculated. Maybe the pair (state, state_objective) never gets used at the same time as (objective, dual_values), but this is hard to formally impose.
Maybe one solution is to update the needs_duals to specialize to the ConicDuality (which does not re-solve the problem). What do you think?
This reverts commit 2eb30feaa18e551010398e55e4a8c59b1fbe8723.
@bfpc I'm reverting #531 because it doesn't account for cases with integer variables (so no dual solution will be present).
The place to fix it is probably here:
https://github.com/odow/SDDP.jl/blob/ed85827e85895ae7f3233d13a8bc4836db4fb56a/src/plugins/duality_handlers.jl#L112-L119
but that can be a separate PR.