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
Is it possible to specify discrete valid values for each dimension of the parameter (of the cost function) such that only those values on the grid are evaluated? A special case would be optimizing a single integer variable. An extension is being able to specify some dimensions of the parameter to be discrete and specify the discrete space over which it takes values. Some problems are discrete in nature, but combinatorics kills any attempt to do greedy grid search due to the curse of dimensionality.
One possible solution would be to put a map inside the evaluated function (logL) s.t. it maps continuous space into discrete one(e.g. p(1)=round(p(1)). But this wastes function evaluations for no good reason, and would be horrible for slow function evaluations.
The text was updated successfully, but these errors were encountered:
Yep, that's the solution that I currently recommend as a quick fix, but it's clearly not ideal.
I had this request before and I have been thinking of implementing integer variables, but not happening soon, it does require some tinkering.
Is it possible to specify discrete valid values for each dimension of the parameter (of the cost function) such that only those values on the grid are evaluated? A special case would be optimizing a single integer variable. An extension is being able to specify some dimensions of the parameter to be discrete and specify the discrete space over which it takes values. Some problems are discrete in nature, but combinatorics kills any attempt to do greedy grid search due to the curse of dimensionality.
One possible solution would be to put a map inside the evaluated function (logL) s.t. it maps continuous space into discrete one(e.g. p(1)=round(p(1)). But this wastes function evaluations for no good reason, and would be horrible for slow function evaluations.
The text was updated successfully, but these errors were encountered: