test_correct_startpoint_usage[34-nlopt] fails

[dweindl]

e.g. https://github.com/ICB-DCM/pyPESTO/actions/runs/11889668047/job/33126701722?pr=1508

Probably related to the recent nlopt release

___________________ test_correct_startpoint_usage[34-nlopt] ____________________

optimizer = ('nlopt', 11)

    def test_correct_startpoint_usage(optimizer):
        """
        Test that the startpoint is correctly used in all optimizers.
        """
        # cma supports x0, but samples from this initial guess, therefore return
        if optimizer == ("cma", ""):
            return
    
        opt = get_optimizer(*optimizer)
        # return if the optimizer knowingly does not support x_guesses
        if not opt.check_x0_support():
            return
    
        # define a problem with an x_guess
        problem = CRProblem(x_guesses=[np.array([0.1, 0.1])]).get_problem()
    
        # run optimization
        result = optimize.minimize(
            problem=problem,
            optimizer=opt,
            n_starts=1,
            progress_bar=False,
            history_options=pypesto.HistoryOptions(trace_record=True),
        )
        # check that the startpoint was used
>       assert problem.x_guesses[0] == pytest.approx(
            result.optimize_result[0].history.get_x_trace(0)
        )
E       assert array([0.1, 0.1]) == approx([0.317...27 ± 3.2e-07])
E         
E         comparison failed. Mismatched elements: 2 / 2:
E         Max absolute difference: 0.2170485381304927
E         Max relative difference: 2.170485381304927
E         Index | Obtained | Expected                    
E         (0,)  | 0.1      | 0.3170485381304927 ± 3.2e-07
E         (1,)  | 0.1      | 0.3170485381304927 ± 3.2e-07

test/optimize/test_optimize.py:668: AssertionError

[dweindl]

optimizer = ('nlopt', 11)

-> nlopt.LN_PRAXIS

I don't see anything directly related to LN_PRAXIS in stevengj/nlopt@v2.8.0...v2.9.0

There is stevengj/nlopt@6a9fbb0, but the associated issue should have been present in 2.8.0 already.

The most relevant change could be stevengj/nlopt@4fa930c.

[dweindl]

optimizer = ('nlopt', 11)

-> nlopt.LN_PRAXIS

I don't see anything directly related to LN_PRAXIS in stevengj/[email protected]

The relevant change is more recent. Tags from from nlopt and nlopt-python don't match exactly.

Tracked here: stevengj/nlopt#575