Improvements to stats stopping criteria verification

[abelsiqueira]

The way that it set_status! and get_status work right now is too verbose.
You have to pass max_eval, max_iter and max_time every time. Probably these should be stored as part of the stats, or as a separate object - this would have to be internal to stats to avoid allocation.

The done = stats.status != :unknown is prone to error. A function check on stats would be better.

[tmigot]

Ok, so if I follow your idea:

• we would store a boolean value for each status in the stats,
• and, have a function

check_status(stats) = stats.status != :unknown

[abelsiqueira]

I am not sure yet. Maybe something like Stopping with no internal computations would make sense?
I just feel like iter and elapsed_time are already being stored inside stats, so maybe it would make sense to have something as part of stats that check for stopping condition. But anyway, very open ended from my part.

The second, yes, I think a stop_criteria_satisfied(stats), for instance, would make the while look nice: while !stop_criteria_satisfied(stats), etc.

[dpo]

I'm not a big fan of that because it hard-codes stopping criteria. We developed the callback mechanism to provide much more freedom in that respect.

[abelsiqueira]

It doesn't prevent using other stopping criteria nor the callback. Also, these are the criteria that we want in our solvers (already being used). Why is it hardcoded?