Diversity picker bug

[stefdoerr]

There is a bug in the diversity picker which practically just returns random conformations.

ANI-Tools/lib/pyaniasetools.py

Lines 535 to 539 in 7efd945

	dm = scispc.distance.pdist(aevs, 'sqeuclidean')
	picker = rdSimDivPickers.MaxMinPicker()
	seed_list = [i for i in range(Ngen)]
	np.random.shuffle(seed_list)
	ids = list(picker.Pick(dm, Ngen, Nkep, firstPicks=list(seed_list[0:5])))

Here you pass the results of pdist to MaxMinPicker. However if you read the RDKit blogpost on that class you will see the funny paragraph

Note that there are some examples of using this approach floating around on the web that calculate the distance matrix in the wrong order. I've done my best to find and either fix or remove them, but there are no doubt still some bad ones out there.

If you look at their code they pass the lower triangular matrix by iterating:

for i in range(1, N):
   for j in range(0, i):
      dists.append(d[i, j])

while if you read the pdist documentation it returns them in order i < j < m so in this form

for i in range(N):
   for j in range(i+1, N):
      dists.append(d[i, j])