-
Added support for keras (#164) and recipes (#179).
-
Moved where
required_pkgs
metadata is stored remotely, from the binary blob to plain text YAML (#176). -
Added an optional renv lockfile that can be stored remotely in model metadata, with a new
check_renv
argument for reading/writing (#154). -
Exposed a new
base_image
argument for creating Dockerfiles (#182).
- The arguments for dealing with a model's input data prototype have changed from using
ptype
to usingprototype
(#166):- In
vetiver_model()
, now usesave_prototype
. - In
vetiver_api()
, now usecheck_prototype
.
- In
-
Added support for k-Prototypes clustering from clustMixType (#163, thanks to @galen-ft).
-
Now vendor renv directly in package (#157).
-
Fixed how
vetiver_ptype()
finds predictors for models (lm()
andglm()
) with interactions (#160). -
New argument added to
vetiver_write_docker()
to pass in additional packages to be installed, such asrequired_pkgs(board)
(#159). -
New function
vetiver_prepare_docker()
creates all necessary files to deploy a basic vetiver model via Docker (#165). -
Fixed a bug in handling all-
NA
columns when predicting on avetiver_endpoint()
(#169).
-
Trailing slashes are now removed from
vetiver_endpoint()
(#134). -
Added support for GAMs from mgcv (#135) and stacks (#138).
-
Added
augment
method for vetiver endpoint (#141). -
Added
apt-get clean
to Dockerfile to reduce container size (#142, thanks to @csgillespie). -
Fixed bug where not all system requirements were added to the Dockerfile (#142, thanks to @csgillespie).
-
Added bundle support for relevant models (#145).
-
Fixed bug in generating Dockerfiles when explicitly requiring the stats package (#147).
-
Now pass the dots for writing a pin through to vetiver allowing, for example,
vetiver_pin_write(b, v, access_type = "all")
on RStudio Connect (#121, #122). -
vetiver_pin_metrics()
now finds thetype
of the existing pin and updates with the same type (#122).
-
The lockfile created by
vetiver_write_docker()
can now be named via the argumentlockfile
, and its default isvetiver_renv.lock
(#100). -
Switched the default for
overwrite
invetiver_pin_metrics()
fromTRUE
toFALSE
. UsingFALSE
is a better choice for interactive use whileTRUE
is probably the right choice for reports or dashboards that are executed on a schedule (#104). -
Added an optional
EXPOSE PORT
line to the generated Dockerfile, to help out Docker Desktop users (#105). -
Added model monitoring dashboard template (#98). To knit the example vetiver monitoring dashboard, execute
vetiver::pin_example_kc_housing_model()
to set up demo model and metrics pins. -
The OpenAPI spec generated for a vetiver model now includes the model version when applicable.
-
Added option to write a Plumber file without packages listed for RStudio Connect purposes (#112).
-
Added new function
vetiver_create_rsconnect_bundle()
as an alternative deployment strategy (#113).
-
Add functions for model monitoring (#92).
-
Update all URLs in package for new documentation sites.
- Improve how Dockerfiles are generated.
-
Update all tests to use redaction for snapshots.
-
Use ranger conditionally in examples/tests.
-
Generate Dockerfiles to deploy model (#71).
-
Added support for glm (#75) and ranger (#76).
-
Deprecated
vetiver_pr_predict()
in favor of usingvetiver_api()
(#77).
-
Added support for tidymodels (#51), caret (#52), and mlr3 (#56).
-
Added vignette.
-
Escalated parsing/conversion warnings to errors in
vetiver_type_convert()
(#60). -
Added
predict
method forvetiver_model()
(in addition tovetiver_endpoint()
) (#61). -
New function
vetiver_deploy_rsconnect()
to deploy to RStudio Connect. -
Added R Markdown template for Model Card for responsible, transparent model reporting (#62, #63).
-
Modularized
vetiver_pr_predict()
to support more advanced use cases (#67).
- Initial CRAN release of package.