DiscreteMutator allows for a non-zero diagonal in transition matrix

[cbennington852]

At the moment, we are allowed to input matrix's with non-zero diagonal's.

And example of this is below.
[
[0, 0.5, 0.5],
[0.1, 0 , 0.9],
[0, 0, 1]
]
This makes little sense, as if this section we selected, the Node would mutate from itself -> itself....

Suggested changes add the following after line 202 in mutators.py

        if len(transition_matrix) != 1:
            for x in range(0 , len(transition_matrix)):
                if transition_matrix[x][x] != 0:
                    raise ValueError(
                        f"transition_matrix diagonal must be zero. "
                    )

[wsdewitt]

Thanks! I think this fix would work, but could be made more concise (and efficient) as:

        if np.diag(transition_matrix).any():
            raise ValueError(f"transition_matrix diagonal must be zero.")

The case len(transition_matrix) == 1 can't occur as long as len(state_space) > 1 (which we should be enforcing because a discrete space with one state has no notion of mutation).

[cbennington852]

sounds good! The changes are stored in the charles-branch.

Ultimately I had to add a edge case otherwise the code would not function. I think this has to to with how the matrix is initialized? Without the (not len(transition_matrix) == 1) the function would error out constantly. Perhaps this is due to line 187 transition_matrix.shape[1] . Which initializes the matrix as a [[1.]]

if (not len(transition_matrix) == 1) and np.diag(transition_matrix).any():
                raise ValueError(f"transition_matrix diagonal must be zero.")

[wsdewitt]

I believe line 187 checks that the row and column dimensions are equal (note the .shape attribute gives the array dimensions, it doesn't initialize). Can you provide details about the error you're seeing?

[cbennington852]

Running this code with some test cases we get

None of my test cases init an array at [[1.]] . Not really sure why this occurs. The test cases are only calling DiscreteMutator a total of two times.

Remvoing the (not len(transition_matrix) == 1) and we get the error .

[cbennington852]

I may have found the issue!

In tree.py we set the DiscreteMutator to be a 1x1 matrix initialized to one. This was likely what was causing this error message.