Aaditya Rangan’s missing value filling method

after filling the missing values with the median value per dimension do a low-rank approximation (svd) and re-project the matrix and repeat until the missing value vector values converge. It was tested to about ~70% missing matrix.

also, he runs a fast-tsne (Mannas’ implementation) followed by K-means to cluster the genetic data. Better discrimination when you apply missing-value filling procedure using the low-rank matrix method.