Nail down that log likelihood calculation

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

Nail down that log likelihood calculation #30

Open TommyJones opened 4 years ago

TommyJones commented 4 years ago

Currently, the log likelihood calculation is the same one used in textmineR. This calculation makes sense to me given the conventional definition of likelihood: probability of the data given the model parameters. However the more widely accepted likelihood calculation incorporates prior information. I'd like to return both likelihoods. The latter has continually been problematic for me to calculate, given that its positive.

TommyJones commented 2 years ago

Another thing that needs fixing: the likelihood calculation assumes that eta is a vector, not a matrix. Similarly related to #58