An approximation of the entries of a nonnegative sparse matrix Y with integer combinations of the latent components of Nonnegative Matrix Factorization (NMF) of Y. Alternating Least Squares (ALS) and Branch-and-Bound approach are used.
Approximation of this type can arise e.g. when a target set of ingredients needs to be delivered optimally through a combination of a fixed maximum of predifined ingredient compounds (number of compunds is less than the number of ingredients). For details, see the Jupyter notebook.