Wrong Numpy Array Type When Using Choice in MixedVariableMating

[Deathn0t]

Hello,

When using the MixedVariableGA with MixedVariableMating (default) with a Choice(options=["const", "cosine"]) variable, I noticed that the following step

            mutation = self.mutation[clazz]
            _off = mutation(_problem, _off, **kwargs)

was generating "cosin" instead of "cosine", then I realised that the X[mut, k] = v affectation withing the _do of ChoiceRandomMutation was transforming "cosine" (correct sampled value) to "cosin".

This helped me understand that X was a numpy array, which was probably set with a default type not compatible with any object (looked like a fixed memory reservation looking at the truncation of "cosine" to "cosin").

Going back to the MixedVariableMating class, the _parents are by default created without dtype, I suggest to use dtype="O" for variables of type Choice (looking at the doc/code, it can be any object in the list passed to Choice(options=[...]):

            _parents = [
                [Individual(X=np.array([parent.X[var] for var in list_of_vars], dtype="O" if clazz is Choice else None)) 
                  for parent in parents] 
                for parents in pop
            ]

This resolved this quite problematic bug in my use of PyMOO.

[blankjul]

Such an easy and elegant fix!

This is an ugly bug I was looking into once before (and does not happen across all Python/Numpy versions and OS) and I was hoping it is already fixed.