Added support for keras (#164), recipes (#179), and luz (#187, @dfalbel).
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, #192).
Exposed a new base_image argument for creating
Dockerfiles (#182).
Added new vetiver_deploy_sagemaker() function plus
vetiver_endpoint_sagemaker() and other needed functionality
for deploying on Amazon SageMaker (#186, @DyfanJones).
Added new additional GET endpoint for model
/metadata (#194).
ptype to using prototype
(#166):
vetiver_model(), now use
save_prototype.vetiver_api(), now use
check_prototype.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() and glm()) with interactions
(#160).
New argument added to vetiver_write_docker() to pass
in additional packages to be installed, such as
required_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 a vetiver_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 the
type of the existing pin and updates with the same type
(#122).
The lockfile created by vetiver_write_docker() can
now be named via the argument lockfile, and its default is
vetiver_renv.lock (#100).
Switched the default for overwrite in
vetiver_pin_metrics() from TRUE to
FALSE. Using FALSE is a better choice for
interactive use while TRUE 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.
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 using
vetiver_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 for
vetiver_model() (in addition to
vetiver_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).