nmf0: Fit a Fast Scalable Non-negative Matrix Factorization using...

In kkdey/NMF0: A Fast Scalable and Flexible Non Negative Matrix Factorization Method

Description

Fits a grade of membership model using non-negative matrix factorization using L-0 penalization that can handle large scale input data and is fast and scalable. Also allows for unsupervised and semi-supervised set ups for factors/clusters. X=LF

Usage

nmf0(Xmat, K = 1L, knownF, lambda1 = 0.1, lambda2 = 0.1,
  lambda3 = 0.1, tol = 1e-04, maxiter = 1000, verb = TRUE,
  init_method = "random", hard_keep = NULL)

Arguments

K	the number of clusters to fit. Must be an integer.
knownF	The number of known factors to be used. Default is missing in which case the unsupervised NMF0 method is used.
lambda1	The tuning parameter for the L-0 penalty on the loading matrix L
lambda2	The tuning parameter for the L-2 penalty on the loading matrix L
lambda3	The tuning parameter for the L-2 penalty on the factor matrix F
tol	The relative tolerance which when met calls for stoppage in the optimization run
maxiter	The maximum number of iterations for which to run the iterative updates in nmf0
verb	If TRUE, prints the progress of the model fit
init_method	The method for initializing the NMF method. Can be one of two values - random and svd. when init_method=random, the initial updates to L and F are decided randomly
hard_keep	The maximum number of clusters that are representative of each sample.
X	The input data matrix with rows being the features and columns the samples.

Value

Outputs the best NMF0 fitted model for cluster K that includes the estimated L and F matrices, and the model likelihood.

References

Hussein Hazimeh and Rahul Mazumder.2018. Fast Best Subset Selection: Coordinate Descent and Local Combinatorial Optimization Algorithms. arXiv preprint arXiv:1803.01454.

Examples

data("ex.counts")
out <- nmf0(t(ex.counts), K=4, tol=0.001, maxiter=1000, verb=FALSE)

kkdey/NMF0 documentation built on Jan. 19, 2021, 10 a.m.