def run_affinity_propagation(affinities):
smin = np.min(affinities)
smax = np.max(affinities)
candidate_preferences = np.linspace(smin, smax, 10)
ap = AffinityPropagation(affinity='precomputed')
for preference in candidate_preferences:
ap.preference = preference
ap.fit(affinities)
indices = ap.cluster_centers_indices_
if indices is not None and len(indices) > 1:
break
return ap
def run_affinity_propagation(self, affinities, leaves):
affinities_list = squareform(affinities)
preference = np.median(affinities_list)
branching_factor = np.inf
while branching_factor > self.max_branching_factor:
ap = AffinityPropagation(affinity='precomputed')
ap.preference = preference
ap.fit(affinities)
if leaves:
branching_factor = 0
else:
branching_factor = max(Counter(ap.labels_).values())
#branching_factor = affinities.shape[0]//len(ap.cluster_centers_indices_)
preference += (np.max(affinities_list) - np.min(affinities_list))/500
print('branching factor = {0}'.format(branching_factor))
print('labels')
print(Counter(ap.labels_))
return ap.cluster_centers_indices_, ap.labels_
