doyouevenstacksmash / data-analysis Goto Github PK

Likelihood function might need a normalization parameter.

Currently we are using both jnp and np for arrays. This has a known effect on precision, and could be the cause of some inconsistencies in the patient traversal.

shows up here

k means empty cluster
https://github.com/DoYouEvenStackSmash/data-analysis/blob/4d42a33719890f4ec7719a1dd8bc50e4dc9e18d2/src/python-processing/kmeans.py#L71C4-L71C4

early stop condition:

Currently it is possible for the number of centroids to decrease during iterations of k means. This is permissible and triggers an early stop condition, which greatly improves performance but possibly at the cost of soundness. more investigation required to determine whether this is important for overall good clustering and tree search.

here

Currently, this behavior is embedded in tree build. It should be extracted to avoid broadcasting errors and semantic issues.

Currently I calculate likelihoods by doing the following:

The paper suggests another which... is a little harder to reason about considering we don't have any kind of temporal ordering in the data as far as I know

Maybe we're supposed to do some numerical technique overall for the posterior integration, it seems the likelihood is incorrect.

What to do?