Fix Ensemble Bug

[PaulJonasJost]

Speedup in history choosing. Now also covers the case in which the max size is smaller then the number of starts, as before this resulted in no ensemble

[codecov-commenter]

Codecov Report

Merging #1063 (515090c) into develop (160c2a8) will decrease coverage by 4.57%.
The diff coverage is 87.73%.

❗ Your organization is not using the GitHub App Integration. As a result you may experience degraded service beginning May 15th. Please install the Github App Integration for your organization. Read more.

@@             Coverage Diff             @@
##           develop    #1063      +/-   ##
===========================================
- Coverage    88.16%   83.59%   -4.57%     
===========================================
  Files           79      147      +68     
  Lines         5257    11389    +6132     
===========================================
+ Hits          4635     9521    +4886     
- Misses         622     1868    +1246     

Impacted Files	Coverage Δ
pypesto/engine/mpi_pool.py	0.00% <0.00%> (ø)
pypesto/objective/amici/__init__.py	100.00% <ø> (ø)
pypesto/objective/amici/amici.py	84.71% <ø> (ø)
pypesto/objective/amici/amici_calculator.py	83.33% <ø> (ø)
pypesto/objective/amici/amici_util.py	71.65% <ø> (ø)
pypesto/objective/base.py	87.62% <ø> (-0.10%)	⬇️
pypesto/objective/finite_difference.py	94.16% <ø> (ø)
pypesto/objective/function.py	94.91% <ø> (+19.91%)	⬆️
pypesto/objective/jax/__init__.py	100.00% <ø> (ø)
pypesto/objective/jax/base.py	91.56% <ø> (ø)
... and 137 more

[dilpath]

👍

[dilpath]

This and the previous line look like they could be combined into

Suggested change

	ens_ind = [np.flatnonzero(fval <= cutoff) for fval in fval_traces]
	# count the number of candidates per start
	n_per_start = np.array([len(start) for start in ens_ind])
	n_theo = np.array([len(start) for start in ens_ind])
	n_theo = np.array([(fval <= cutoff).sum() for fval in fval_traces)])

[m-philipps]

I'm not sure if this is already the desired behaviour in get_vector_indices from line 1137, especially when not using a cutoff.

When distribute = True but n_vectors is 1 or 2, the optimization startpoint or start- & endpoint are used, respectively.

When distribute = False the first n_vector points will be used, i.e. the startpoint and first steps.