Closed emorice closed 2 years ago
Cancelling. This was the issue that prompted the writing of GEMZ, but the first examples uncovered other interesting phenomena (regularization in high dim, mixture models in high dim) that drove the development of quite a few models and a lot of tooling.
This includes a higher-level way to code low-rank updates in gemz.linalg
, meaning that writing models with corrections for means is not really an issue anymore. See models.gmm
and models.igmm
for example of models that handle rank-2 updates due to loo-means with much simpler code than CMK. As such, writing a new variant of CMK with cluster means is not scheduled anymore.