delete_nan_inf can handle list input type

[m-philipps]

• fix Parameters plot error when result has None #1107
• do not attempt to convert parameter list with None to numpy array
• filter out None from parameter vector within delete_nan_inf
• add a test that covers optimizer result with x=None

[codecov-commenter]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (7237234) 83.87% compared to head (cf1bbcb) 83.87%.

Files	Patch %	Lines
pypesto/util.py	85.71%	1 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #1243   +/-   ##
========================================
  Coverage    83.87%   83.87%           
========================================
  Files          148      148           
  Lines        12067    12071    +4     
========================================
+ Hits         10121    10125    +4     
  Misses        1946     1946           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[PaulJonasJost]

Why is making that specifically a list within a list required?

[m-philipps]

This is making the use of delete_nan_inf consistent.
The values are given over to delete_nan_inf in place of the optimization result parameters. These are parameter vectors within a list, i.e. one list (or None) per start.

[m-philipps]

Does anyone know where this condition is needed? It seems counter-intuitive because the function deletes nan values from the given list.

pyPESTO/pypesto/util.py

Lines 282 to 296 in c012c19

	else:
	# If none of the fvals are finite we try to construct an
	# empty np.ndarray with the correct dimension (other functions rely
	# on x.shape[1] to be of correct dimension)
	x = np.array(x)
	x = np.empty(
	(
	0,
	x.shape[1]
	if x.ndim == 2
	else x[0].shape[0]
	if x[0] is not None
	else xdim,
	)
	)

[dweindl]

Does anyone know where this condition is needed? It seems counter-intuitive because the function deletes nan values from the given list.

No idea where it's needed. Yes, it's deletes the nans, but it makes sure that the dimensions of x still match (an np.array with shape (0, 0) is different from (0, 2), although they are both empty).

[Doresic]

Moving the delete_nan_inf to here makes it so the plotting of reference points doesn't go through it. Is that an issue? I've never used reference points, but they also possibly might contain infinite/nan values, no?

[m-philipps]

Good point, it's not a problem though, nans are just not shown in the line plot.

[stephanmg]

Looks fine, branch needs update.

[Doresic]

Looks good! Thank you