Recursively check for Evaluate() in MOO

[jonpsy]

Refer #282 Section II.

Is there a strong reason for having template<MatType, FunctionType> instead of template<FunctionType, MatType>, the problem is when we do recursive call, I need the FunctionType to be the at the end for it to work. I don't see any other way.

Infact if we go by template<MatType, FunctionType>, I can integrate CheckMultiArbitraryFunctionTypeAPI() into CheckArbitraryFunctionTypeAPI(), and it will work like normal + MOO checks.

[jonpsy]

I think template specialization is the only way out of it. Thoughts? @zoq

[zoq]

I don't mind the solution, I think it's fine; it would also be fine for me to change the order, it's an internal function so we will not break anything. @rcurtin what do you think?

[jonpsy]

I found two solutions for the problem, one is:

 for(....)
{
    using FunctionType = typename std::tuple_element<
        I, std::tuple<ArbitraryFunctionType...>>::type;
    traits::CheckArbitraryFunctionTypeAPI<
        FunctionType, MatType>();
....
}

Inside Evaluate objectives(see nsga2_impl.hpp in current PR)

OR

Just take all the arguments, and keep the MatType as separate (as is implemented in the current PR).

[jonpsy]

• Everything happens at compile time.
• Uses the exact same interface.
• Automagically handles a higher number of functions.

Perfect!

Side Note: Static Code Analysis is falling because "it can't find cmake". I don't think it has anything to do with the code itself (I may be wrong).

[zoq]

@mlpack-jenkins test this please

[jonpsy]

Trying this on NSGA2 is giving me some errors, I'll fix it and ping you back.

Note: The issue is fixed. Thanks to that error, the code is even better. ping: @zoq

[zoq]

Would be nice if you could add a MMO function to:

https://github.com/mlpack/ensmallen/blob/master/tests/function_test.cpp

[zoq]

Would be nice if you could add a MMO function to:

https://github.com/mlpack/ensmallen/blob/master/tests/function_test.cpp

Nevermind, it's another test suite.

[zoq]

Looks great to me, thanks for putting this together 👍

[zoq]

@mlpack-jenkins test this please

[jonpsy]

@zoq I suppose I need to rebase ?

[zoq]

@zoq I suppose I need to rebase ?

No, I just like to see if the mlpack master build test works as expected.

[mlpack-bot]

Second approval provided automatically after 24 hours. 👍

[jonpsy]

mlpack master build fails:

g.Component(0).Mean()[0] == Approx(0.5).epsilon(0.05)

FAILED:
  REQUIRE( g.Component(0).Mean()[0] == Approx(0.5).epsilon(0.05) )
with expansion:
  0.5299031563 == Approx( 0.5 )
at /home/jenkins/workspace/pull-requests-ensmallen-mlpack-build/mlpack/src/mlpack/tests/gmm_test.cpp:276

Is this expected?

[zoq]

mlpack master build fails:

g.Component(0).Mean()[0] == Approx(0.5).epsilon(0.05)

FAILED:
  REQUIRE( g.Component(0).Mean()[0] == Approx(0.5).epsilon(0.05) )
with expansion:
  0.5299031563 == Approx( 0.5 )
at /home/jenkins/workspace/pull-requests-ensmallen-mlpack-build/mlpack/src/mlpack/tests/gmm_test.cpp:276

Is this expected?

Some mlpack tests are non-deterministic, so they can fail sometimes, in most cases a simple threshold adjustment is all what is needed.

[zoq]

Thanks again for the contribution.