Elements of Statistical Learning has this on such NMF loss function (section 14.6 Non-negative Matrix Factorization):

The matrices $\mathbf{W}$ and $\mathbf{H}$ are found by maximizing $$ L(\mathbf{W}, \mathbf{H}) = \sum_{i=1}^N\sum_{j=1}^p[x_{ij} \log(\mathbf{WH})_{ij} − (\mathbf{WH})_{ij} ]. \; \; \; \; \; \; \; \; \; (14.73) $$This is the log-likelihood from a model in which $x_{ij}$ has a Poisson distribution with mean $(\mathbf{WH})_{ij}$— quite reasonable for positive data.

What if $x_{ij}$ is not a natural number? Is there a class of distributions that generalizes Poisson distribution that has such log-likelihood (up to a term that doesn't depend on distribution mean)?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.