NonNegativeMatrixFactorization Class

Use the NonNegativeMatrixFactorization class to compute an approximate factorization of a Matrix into matrices with only positive components. In other words, it rewrites a matrix X as a product

X = WH

where W and H have only zero or positive components. The inner dimension of the product (the number of columns of W and the number of rows of H) may be less than the rank of X. In this case, the factorization is only an approximation to the original matrix. This is actually very common, since the Non-Negative Matrix Factorization is often used to reduce the dimensionality of data.

The constructor NonNegativeMatrixFactorization takes two arguments. The first is the matrix that is to be factorized. The second is the inner dimension of the product. The Decompose()()() method performs the actual calculation using an alternating least squares method. The resulting factors are then available as the LeftFactor and [!:RightFactor] properties.