tidymodels
step_poisson_matric_factorization()
I've also been looking at similar methods for binary data. Some references:
- A logistic non-negative matrix factorization approach to binary data sets
- Probabilistic visualisation of high-dimensional binary data
- Probabilistic principal component analysis
- A Generalized Linear Model for Principal Component Analysis of Binary Data
I'm sure that there are some R packages for at least one of these.
I read about Gamma-Poisson factorization for single categorical columns in Patricio Cerda, Gaël Varoquaux. Encoding high-cardinality string categorical variables. 2019. (analogous to solving the following non-negative matrix factorization (NMF) with the generalized Kullback-Leibler divergence)
The paper includes an interesting online algorithm.
Does it make sense to use {reticulate} with this python implementation? https://skrub-data.org/stable/reference/generated/skrub.GapEncoder.html
without having looked at the documentation, i lean on the side of translating the method to R rather than using {reticulate}. Purely on the basis of developer burden. Using {reticulate} in a package is already not the best experience, and then you have to worry about breaking changes from the python implementation etc etc.