-
Notifications
You must be signed in to change notification settings - Fork 105
Description
This was discussed in two previous issues, #35 and #953. The new model is being added in this fork branch, and an extension with engines is underway here.
First see this article for guidance on adding a new model.
Checklist for adding a new model
Copied from this article:
-
Create the main modeling function (e.g.
nearest_neighbors()
). When choosing argument names, check for existing ones across other model types which could be reused, e.g.,trees
for the number of trees is used across different tree-based models likerand_forest()
andboost_tree()
.- here
- Model parameters currently include
penalty
,mixture
,ordinal_link
, andodds_link
.
-
Register the model and a mode in the parsnip model “database” with
parsnip::set_new_model()
andparsnip::set_model_mode()
. [example] -
Create an
update()
method. -
Add parameter objects for main parameters to dials, if needed. See How to create a tuning parameter function for more details.
- here
- This is done in a coordinated fork branch.
-
Write unit tests for the basic modeling function (independent of the engines).
-
Add a pkgdown entry.
-
Add model description to the internal
model_descs
tibble. -
Run
purrr::map(parsnip:::extensions(), ~ library(.x, character.only = TRUE))
. -
Run
parsnip:::update_model_info_file()
. -
Make sure that no entries are removed from
inst/models.tsv
. -
Add a documentation entry in
NEWS.md
.