pypesto.profile.profile_next_guess.get_reg_polynomial correct?

[dweindl]

Hi, can somebody comment on whether the following code makes sense?

This is polynomial regression to extrapolate parameter values for parameter profiles in the adaptive stepsize case.

pyPESTO/pypesto/profile/profile_next_guess.py

Lines 355 to 372 in 223859e

	# Do polynomial interpolation of profile path
	# Determine rank of polynomial interpolation
	regression_tmp = np.polyfit(
	current_profile.x_path[par_index, -1:-reg_points:-1],
	current_profile.x_path[i_par, -1:-reg_points:-1],
	reg_order,
	full=True,
	)
	# Decrease rank if interpolation problem is ill-conditioned
	if regression_tmp[2] < reg_order:
	reg_order = regression_tmp[2]
	regression_tmp = np.polyfit(
	current_profile.x_path[par_index, -reg_points:-1],
	current_profile.x_path[i_par, -reg_points:-1],
	int(reg_order),
	full=True,
	)

In the ill-conditioned case, different parameter vectors are used than in the initial attempt. They don't include the most recent parameters. This seems wrong to me, but maybe somebody knows better.

EDIT: Well, actually both cases seem to be wrong. Both use only reg_points - 1 points instead of the presumably intended reg_points. In the first case, the oldest one is omitted, in the second case the most recent one is omitted.

What I think it should look like: #1226

[PaulJonasJost]

agreed, this should be changed 👍🏼