search

TommyJones / tidylda

Implements an algorithim for Latent Dirichlet Allocation using style conventions from the [tidyverse](https://style.tidyverse.org/) and [tidymodels](https://tidymodels.github.io/model-implementation-principles/index.html).
Other
41 stars 3 forks source link

Gibbs sampler makes erroneous assumption about magnitude of rows of eta #58

Closed TommyJones closed 3 years ago

TommyJones commented 3 years ago

This comment in #8 points out that the current Gibbs sampler assumes each row of eta has the same magnitude. Changes to how refit happens as addressed by #49 require that rows of eta have different magnitudes. The Gibbs sampler needs to be updated to have different magnitudes (eta_sum) for each row.

TommyJones commented 3 years ago

https://github.com/TommyJones/tidylda/commit/f184b88df9f469d264b9d66676471e34ede48f2a

Note that there's still an issue with the likelihood calculation as documented in #8