Computes low dim embedding, constructs KNN graph on the embedding -> unweighted adjacency Calls manifold learning algorithm which uses the normalized sample vectors and the unweighted adjacency matrix to compute a low rank approximation of the data.
SimilarityM( lambda = 0.5, data, comps_knn = NULL, use_umap_indices = FALSE, pre_embed_method = "umap", ... )
lambda | the balance term between the rank of Z and the error, default is 0.5 |
---|---|
data | the expression data, where each column is treated as a normalized vector |
comps_knn | number of components to use for knn, overrides eigengap-based inference |
use_umap_indices | use the knn indices computed during umap embedding to impose sparsity on L2R2, instead of recomputing based on the layout. |
pre_embed_method | how the initial non-linear embedding is performed, default is 'umap' |
... | extra arguments passed to umap or Rtsne |
a list containing
the similarity matrix
the error of the ADMM step
the KNN sparsity constraint is based on this embedding