Wishlist: Feature names and effects for wordshoal

kbenoit commented 6 years ago

Issue by methodds Tuesday Mar 14, 2017 at 16:01 GMT Originally opened as https://github.com/kbenoit/quanteda/issues/598

Hi,

as another suggestion for improvements, it would be nice if wordshoal models could include feature names and effects. These would be the wordshoal "equivalents" of @features, @beta, @psi in textmodel_wordfish(). I cannot find this in the documentation:

tol: log-posterior tolerance used in fitting
dir: global identification of the dimension
theta: estimated document positions
beta: debate marginal effects
alpha: estimated document fixed effects
psi: estimated document debate-level positions
groups: document groups
authors: document authors
ll: log likelihood at convergence
se.theta: standard errors for theta-hats
data: corpus to which the model was fit

kbenoit commented 6 years ago

Comment by methodds Wednesday Mar 15, 2017 at 18:22 GMT

After reading through Lauderdale and Herzog (2016) again, this would be the word names and loadings as described at the bottom of this table:

cschwem2er commented 6 years ago

Thanks for still working on wordshoal, even though its not officially in quanteda anymore. Unfortunately I still haven't figured out how to extract the word loadings. Is this possible with the current version?

amatsuo commented 6 years ago

I understand the importance of this statistic. I will implement it along with the debate loading.

kbenoit / wordshoal

Wishlist: Feature names and effects for wordshoal #3