ExaScience / smurff

Bayesian Factorization with Side Information in C++ with Python wrapper
MIT License
70 stars 14 forks source link

Does SMURFF implement a subset of Macau? #127

Closed suchow closed 4 years ago

suchow commented 4 years ago

My understanding from the original Macau paper is that the algorithm is defined over a hypergraph of nodes and relations, jointly factorizing the tensors and producing a single set of latents for each entity. However, SMURFF appears to allow only two or three entities, factoring a single matrix or tensor (plus side-information for each entity).

Is it possible, for example, to use SMURFF to jointly factor a model with several entities and relations between them?

tvandera commented 4 years ago

Hi Jordan,

ExaScience/smurff and jaak-s/macau are fast, but only support tensor/matrix factorization with optional side info.

The julia implementation of Macau (https://github.com/jaak-s/BayesianDataFusion.jl) supports multi-relation factorization.

suchow commented 4 years ago

Thanks!