like in https://scikit-learn.org/0.15/modules/generated/sklearn.neighbors.kneighbors_graph.html#sklearn.neighbors.kneighbors_graph * it should be easy for cosine or shadowsimilarity : - X = X.normalize(axis=1, norm="l2") - mask = 1 - X.dot(X.transpose(), top=k+1) - mask = mask.without_diagonal()
