Reorder changes phenotype for individuals with parameters

[jakobj]

Since parameter values are stored according to their position in the cartesian graph, reordering nodes without updating the (position, value) pairs will change the phenotype of individuals.

Here is a modified version of the reorder test that reproduces the bug:

def test_genome_reordering_empirically(rng):
    # empirically test that reordering does not change the output function of a genome

    pytest.importorskip("sympy")

    genome_params = {
        "n_inputs": 2,
        "n_outputs": 1,
        "n_columns": 14,
        "n_rows": 1,
        "levels_back": None,
        "primitives": (cgp.Mul, cgp.Sub, cgp.Add, cgp.ConstantFloat, cgp.Parameter),
    }

    genome = cgp.Genome(**genome_params)

    # f(x_0, x_1) = x_0 ** 2 - x_1 + 0.5
    dna_fixed = [
        ID_INPUT_NODE,  # x_1 (address 0)
        ID_NON_CODING_GENE,
        ID_NON_CODING_GENE,
        ID_INPUT_NODE,  # x_1 (address 1)
        ID_NON_CODING_GENE,
        ID_NON_CODING_GENE,
        0,  # Mul ->  x_0^2 (address 2)
        0,  # x_0
        0,  # x_0
        1,  # Sub -> x_0^2 - x_1 (address 3)
        2,  # x_0^2
        1,  # x_1
        1,  # Sub -> 0 (address 4)
        0,  # x_0
        0,  # x_0
        3,  # const -> 1 (address 5)
        2,
        3,
        3,  # const -> 1 (address 6)
        0,
        0,
        4,  # param -> 0.9 (address 7)
        4,
        3,
        4,  # param -> 0.4 (address 8)
        7,
        2,
        2,  # Add -> x_0^2 - x_1 + 1 (address 9)
        3,  # x_0^2 - x_1
        5,  # 1
        1,  # Sub -> x_0^2 - x_1 + 1 - 0.9 (address 10)
        9,  # x_0^2 - x_1 + 1
        7,  # 0.9
        2,  # Add -> x_0^2 - x_1 + 1 - 0.9 + 0.4 (address 11)
        10,  # x_0^2 - x_1 + 1 - 0.9
        8,  # 0.4
        3,  # const (address 8)
        0,
        1,
        3,  # const (address 9)
        0,
        1,
        3,  # const (address 10)
        0,
        1,
        3,  # const (address 11)
        0,
        1,
        ID_OUTPUT_NODE,  # out -> x_0^2 - x_1 + 0.5
        11,
        ID_NON_CODING_GENE,
    ]

    genome.dna = dna_fixed
    genome._parameter_names_to_values["<p7>"] = 0.9
    genome._parameter_names_to_values["<p8>"] = 0.4

    sympy_expression = cgp.CartesianGraph(genome).to_sympy()
    n_reorderings = 100
    for _ in range(n_reorderings):
        genome.reorder(rng)
        sympy_expression_after_reorder = cgp.CartesianGraph(genome).to_sympy()
        assert sympy_expression_after_reorder == sympy_expression

@HenrikMettler can you please take a look at this, you're most familiar with the reorder implementation. I guess one would need to go through the _parameter_names_to_values of the Genome class after reordering to update the parameter names to reflect the new node positions.

once this is fixed we should consider a bugfix release, 0.2.1 @mschmidt87 do you want to take care of this again?