csiro-hydroinformatics / metaheuristics Goto Github PK

It should have used an interface and not depend on the log4net package directly.

Consider giving a spin along the line of Hanselman's post

msbuild on Appveyor :

 error CS0246: The type or namespace name 'MPI' could not be found (are you missing a using directive or an assembly reference?) [C:\projects\metaheuristics\CSIRO.Metaheuristics.Parallel\CSIRO.Metaheuristics.Parallel.csproj]
with 
Commit: 3767eefea82e65fb131793877e39f708f291f0f2 [3767eef]
Date: Sunday, 24 March 2019 12:28:55 PM

This occurs at a time where MPI.NET nuget has been just uploaded at 1.4 and is the first case where this targetts the netstandard2.0 profile. Metaheursitics still targets 4.7.2 but my understanding is that it should be fine to rely MPI.NET. I certainly have no issue on my local windows machine.

Use case: SCE calibration with 5 complexes, but on a laptop with e.g. 4 cores. How to set a max level of threading of 3 to keep the machine responsive. Requested by DR.

I think the cancellation of the algorithm has been messed up with. There should be a way to cancel (rather, stop) the algorithm as soon as possible, be it on user intervention or termination criteria, in a thread-safe manner. Relates to https://jira.csiro.au/browse/WRAA-755.

    private static List<ObjectiveScore> Convert(IObjectiveScores scores)

scores can be null if the termination criterion of teh optimisation was such that for instance not all population init happened. Maybe null values should not be permitted as output from optimisation.

This very likely looks like an issue with Mono. Log an issue with Xamarin. Repro at ~/tmp/SomeConsoleApp/Program.cs

On completion, the Calibration Runner should display the history of objective function & metaparameter values used during the optimisation process.
The calibration history is working for SCE and SCE & Rosenbrock but the other optimisation algorithms only display the best (or last?) optimisation result.

The returned results from evolve() call are an array and it makes sense to return all the results tried ordered so the first one is the best (and for backward compatibility)

proposed fix in
#10