Julian Hall comments

Results 551 comments of


                                            Julian Hall

Bug in python callbacks user_solution and mip_solution

> Basically, since the callbacks are C style, there's some potential issues with memory safety. I have a working solution to clean it all up... but has a minor breaking...

Bug in python callbacks user_solution and mip_solution

The changes made in #2278 appear to have triggered #2400 Perhaps we need more general ways of getting the components of `data_out` @mathgeekcoder? Since @odow will also be in Edinburgh...

Compute dual objective value

Closed by #1984

Take into account discreteness for no cost zero variables in bounds section

We've not forgotten about this: just rather busy ATM

Suggestion: enum for some integer options

Clearly we could make them enums - although enum classes would have downsides - but I'm wary of encouraging people to experiment. Serial dual simplex is much the most robust,...

When primal simplex is optimal with bound shifts, avoid primal corrections when recomputing the primal values: return and use dual simplex to clean up

Thanks, I'll look at this.

When primal simplex is optimal with bound shifts, avoid primal corrections when recomputing the primal values: return and use dual simplex to clean up

It's a false positive, but gives me something to think about. Was there a particular reason why you compiled with Debug or RelWithDebug?

When primal simplex is optimal with bound shifts, avoid primal corrections when recomputing the primal values: return and use dual simplex to clean up

Note for @jajhall The complementarity violation is due to a nonbasic variable being off its bound. Specifically, row 5 is nonbasic, and has (primal / dual) residual (2.27295e-05 / 0.0117425)...

When primal simplex is optimal with bound shifts, avoid primal corrections when recomputing the primal values: return and use dual simplex to clean up

Note for @jajhall In `HEkkPrimal::rebuild()`, `correctPrimal()` is always called in Phase 2. This can shift bounds to remain in Phase 2. However, when dual feasible - ie possibly optimal -...

When primal simplex is optimal with bound shifts, avoid primal corrections when recomputing the primal values: return and use dual simplex to clean up

Note for @jajhall So `options_.simplex_strategy` is restored (to 0). Have to track down when it's set to 4