o .wi @sxddlmZddlZddlmZdedededeeeffdd Zd ed edefd d ZddedededefddZdS)) combinationsN)Tensordatalabelspreturncs|jdd\}fddtt|D}dd|D}tjjtjddt|dDdd d d }tfd dt||D}||fS) aTUpdate and return variables required to compute the Dunn index. Args: data: feature vectors of shape (n_samples, n_features) labels: cluster labels p: p-norm (distance metric) Returns: intercluster_distance: intercluster distances max_intracluster_distance: max intracluster distances T)return_inversecsg|]}|kqSr ).0 label_idx)rinverse_indicesr j/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/torchmetrics/functional/clustering/dunn_index.py "z&_dunn_index_update..cSsg|]}|jddqS)rdim)mean)r cr r r r#rcSsg|]\}}||qSr r )r abr r r r&rrrordrcs*g|]\}}tjj||ddqS)rr)torchlinalgnormmax)r cimu)rr r r)s) uniquerangelenrrrstackrzip)rrr unique_labelsclusters centroidsintercluster_distancemax_intracluster_distancer )rr rr _dunn_index_updates  r*r(r)cCs||S)zCompute the Dunn index based on updated state. Args: intercluster_distance: intercluster distances max_intracluster_distance: max intracluster distances Returns: scalar tensor with the dunn index )minr)r(r)r r r _dunn_index_compute0s r,rcCst|||\}}t||S)aCompute the Dunn index. Args: data: feature vectors labels: cluster labels p: p-norm used for distance metric Returns: scalar tensor with the dunn index Example: >>> from torchmetrics.functional.clustering import dunn_index >>> data = torch.tensor([[0, 0], [0.5, 0], [1, 0], [0.5, 1]]) >>> labels = torch.tensor([0, 0, 0, 1]) >>> dunn_index(data, labels) tensor(2.) )r*r,)rrrpairwise_distance max_distancer r r dunn_index>s r/)r) itertoolsrrrfloattupler*r,r/r r r r s  "