Open koheiw opened 4 years ago
That’s a fine idea, and whatever we can do to make it extensible or more general-purpose is something I’d like to develop. One though is to provide a way for all textmodel_*()
functions to work on any sparse matrix, either from Matrix or slam, of course in addition to a dfm. Another is via integrating the predictive models with caret (or tidymodels/hardhat) as in quanteda/quanteda.textmodels#8.
Point me to the code in LSS and newsmap and we can start with yours.
It extends textplot_scale1d
:
https://github.com/koheiw/LSS/blob/e8fad83759ad3d9647f19be7edc6715c95fc00cb/R/textplot.R#L74-L95
Also, as.coefficients_textmodel()
:
https://github.com/koheiw/LSS/blob/e8fad83759ad3d9647f19be7edc6715c95fc00cb/R/textmodel.R#L242
The package no longer builds with quaneda v2.0. I am planing to submit this to CRAN sometime this year.
It is already on CRAN. Not affected much because it does a lot of DYI, but it should use more functions made available in quaneda. https://github.com/koheiw/newsmap/blob/master/R/textmodel_newsmap.R
My LSS and newsmap package uses textmodel methods such as
as.coefficients_textmodel()
andtextplot_scale1d()
, but I do not want to add quanteda.textmodels to their dependency. How about bringing back these functions to quanteda? This helps other people to write their own textmodel's based on our package.