silhouette_score_adj

# def silhouette_score_adj(distances, n_clusters, labels, nearest_other_cluster_mode='min'):
#     for i in range(n_clusters):
#         mask_i = labels==i
#         n_i = mask_i.sum()
#         assert n_i > 0
#         # intra-cluster distance for each sample in i-cluster
#         a = distances[mask_i][:,mask_i].reshape(n_i,n_i).mean(0)

#         other_cluster_numbers = list(range(n_clusters))
#         other_cluster_numbers.remove(i)
#         if nearest_other_cluster_mode == 'mean':
#             bs = []
#             for j in other_cluster_numbers:
#                 mask_j = labels==j
#                 n_j = mask_j.sum()
#                 assert n_j > 0
#                 b = distances[mask_i][:,mask_j].reshape(n_i, n_j).mean(1)
#                 bs.append(b)
#             bs = np.stack(bs, axis=1).min(1)
#         else:
#             bs = distances[mask_i][:,~mask_i].reshape(n_i, -1).min(1)
#         s = (bs - a) / (np.stack([bs,a], axis=1).max(1))
#         return s.mean()