pypesto.profile.parameter_profile incorrectly assumes symmetric bounds

[dweindl]

pypesto.profile.parameter_profile incorrectly assumes symmetric bounds (-lb = ub) here:

pyPESTO/pypesto/profile/walk_along_profile.py

Lines 64 to 75 in 883138a

	# check if the next profile point needs to be computed
	if options.whole_path:
	stop_profile = (
	x_now[i_par] * par_direction >= problem.ub_full[[i_par]]
	)
	else:
	stop_profile = (
	x_now[i_par] * par_direction >= problem.ub_full[[i_par]]
	) or (current_profile.ratio_path[-1] < options.ratio_min)
	if stop_profile:
	break

This causes major problems in case this is not fulfilled:

• ub < -lb: premature stopping, cropped profiles
• ub > -lb: infinite loop

Not sure how this could go unnoticed...

To reproduce:

def test_broken_profiles():
    import pypesto
    from pypesto import profile
    from ..util import rosen_for_sensi

    obj = rosen_for_sensi(max_sensi_order=1)['obj']

    # for infinite loop
    lb = 6 * np.ones(5)
    ub = 10 * np.ones(5)

    # for cropped profile
    lb = -4 * np.ones(5)
    ub = 1 * np.ones(5)

    problem = pypesto.Problem(
        objective=obj,
        lb=lb,
        ub=ub,
    )

    optimizer = optimize.ScipyOptimizer(options={'maxiter': 10})
    result = optimize.minimize(
        problem=problem,
        optimizer=optimizer,
        n_starts=2,
        progress_bar=False,
    )
    profile.parameter_profile(
        problem=problem,
        result=result,
        optimizer=optimizer,
        next_guess_method='fixed_step',
        profile_index=[1],
        progress_bar=False,
        profile_options=profile.ProfileOptions(
            min_step_size=0.5,
            delta_ratio_max=0.05,
            default_step_size=0.5,
            ratio_min=0.01,
            whole_path=True,
        )
    )
    x_path = result.profile_result.list[0][1]['x_path'][1]

    print(x_path.min(), x_path.max())
    assert x_path.max() == ub[1]
    assert x_path.min() == lb[1]

[dweindl]

Introduced in #1014. Shame on the reviewer.