R Observatory: Metrics for parsnip version 1.6.1

Balamuta, James Joseph

doi:10.5281/zenodo.21843040

parsnip

1.6.1Package record

A Common API to Modeling and Analysis Functions

58packages dependsuggested by 50

506.7Kdownloads / year

69.6%test coverage

13/13flavors OK

Overview

About

Maintained by Max KuhnCRAN page ↗GitHub ↗Website ↗Issues ↗Declared funder: Posit Software, PBC

A common interface is provided to allow users to specify a model without having to remember the different argument names across different functions or computational engines (e.g. 'R', 'Spark', 'Stan', 'H2O', etc).

Changes in 1.6.1

Breaking Change

Two translate() helper functions for glmnet are generalized and used to de-duplicate code for ordinalNet and glmnetcr. The latter two no longer silently modify penalty path-governing engine arguments (@corybrunson, #1412 & #1424), which will impact code that relied on these modifications.
logistic_reg() now errors when the outcome has more than two levels, instead of warning. Most engines cannot fit this: glm collapses every level after the first into the event and keras3 returns a column per level, but in both cases the probability columns parsnip labels do not mean what their names say, and glmnet, stan, and brulee error inside the engine. The LiblineaR engine is exempt, since it genuinely fits a multiclass model and multinom_reg() has no LiblineaR engine. The warning was added in 1.1.0 (#545); erroring reflects that for most engines the result was not usable (#1444).

New Features

New model specifications tabular_auto_int(), tabular_chronos(), tabular_icl(), tabular_pfn(), tabular_resnet(), tabular_rln(), and tabular_saint() were added for tabular deep-learning and foundation models, with engines provided by the tabby extension package (#1386).
fit() and fit_xy() have less per-fit overhead, making small or repeated fits (such as during tuning) faster (#1071).
null_model() now supports quantile regression mode, where fitting computes the requested empirical quantiles of the outcome.
For censored regression models, the censoring weights can now be added to the predictions of survival probability by setting add_censoring_weights = TRUE in predict(type = "survival") (#1371).
ordinal_reg() gains arguments threshold_structure and parallel_reg to control threshold constraints and the parallel regression assumption. The ordinalNet engine can use parallel_reg while the clm and vglm engines can use both new arguments (#1393, @corybrunson).

Bug Fixes

The deprecated quantile argument now reaches its deprecation warning when passed via predict(type = "quantile") instead of being rejected as an unknown argument. The error for unknown arguments passed to predict() now lists the offending argument names. (@bjornkallerud, #1258)
Several internal list accesses that resolved only through $ partial matching were corrected: the outcome levels recorded when fitting from a formula, the "ranger" prediction post-processor, and the multi_predict() generic's check for a failed fit. Results are unchanged, but each would have silently started reading a different element had one with a shorter name been added.

Show all changes in 1.6.1Hide the rest of the changes in 1.6.1

Model and engine arguments are now evaluated while the model call is assembled, so the call an engine records no longer contains quosures. This fixes mars() fits with the earth engine when prune_method = "cv" is combined with prod_degree (#432), and set_engine("glmnet", relax = TRUE) for every glmnet engine, which also covers the poissonreg and censored wrappers (#1069). Both engines re-evaluate their own recorded call with base eval(), which cannot handle quosures. A fitted object's $fit$call now shows values such as degree = 2 rather than degree = ~2.
bart() classification fits with the "dbarts" engine now return each observation's own confidence and prediction interval bounds. The bounds were sorted across observations, so each row received some other row's rank-matched limits. Regression intervals were unaffected (#1407).
bart() classification with the "dbarts" engine now errors when the outcome does not have exactly two levels. dbarts models a binary outcome, but parsnip passed it the factor codes without checking, so a third level became a y value of 2 and the returned "probabilities" were not on [0, 1]. This also affected a two-level outcome that retained an unused level: class and probability predictions were silently mislabelled, and interval predictions errored inside rlang::set_names(). The new error names the unused levels and suggests droplevels() when that is the problem (#1444).
boost_tree() with the "xgboost" engine now warns once per session when monotone_constraints is supplied for binary classification. The signs of the constraints are relative to the event level, which is the first factor level unless event_level = "second" is set, so monotone_constraints = 1 constrains the probability of that level rather than of the second one. The engine documentation now describes the convention. Fitted models are unchanged (#796).
boost_tree() models fit with the "xgboost" engine and a function-valued objective now error informatively for predict(type = "class") and predict(type = "prob") instead of returning raw margins labelled as probabilities. xgboost cannot report whether a custom objective produces margins or probabilities, so parsnip cannot convert them; use predict(type = "raw") and apply the matching inverse link yourself, or register a custom engine that post-processes the predictions. Fitting, type = "raw", and regression are unaffected (#999).
fit() and fit_xy() now error when extra arguments are passed through .... The documentation always said these were ignored, but depending on the combination of user and engine interface they were silently dropped, silently applied (a subset argument really did subset the training data on the formula-to-xy path), or raised an internal "unused argument" error. Pass engine arguments to set_engine() and case weights to the case_weights argument. Passing offset gets specific advice, since the general suggestion does not work for it: use offset() in the formula, or pass the offset vector to set_engine() when using fit_xy() (#492, #1439).
mars() classification fits with the "earth" engine now return correct predict(type = "class") results for outcomes with three or more levels. A binary threshold rule was applied regardless of the number of levels, so every multiclass prediction was wrong and the last level could never be predicted. Binary outcomes are unaffected (#472, #1409).
multi_predict() for glmnet engine fits now passes type = "raw" through to glmnet with no post-processing. It was silently ignored for linear_reg(), which returned the usual nested .pred tibble, and errored unhelpfully for logistic_reg() and multinom_reg(). The result is glmnet's own prediction object — a matrix with one column per penalty, or a three-dimensional array for multinom_reg() — rather than a tibble (#857).
The per-type glmnet prediction methods (predict_numeric._elnet(), predict_class._lognet() and the six others like them) were removed. Each only evaluated the model specification before handing off to the corresponding model_fit() method, which the predict() path already does beforehand. Predictions from glmnet models are unchanged. A custom model that carries one of glmnet's fitted classes, such as _elnet, now reaches parsnip's model_fit() methods for these prediction types rather than the glmnet ones (#878).
multi_predict_args() and has_multi_predict() work again for fitted workflows, returning the submodel argument names and TRUE instead of NULL and FALSE. They read from an outdated internal workflows structure. Both now error informatively on an untrained workflow rather than silently reporting that it has no submodel arguments (#1410).
predict_raw() no longer errors when opts contains an argument name that collides with a protected prediction argument such as newdata or object. The colliding entry is now dropped with a warning; previously every path through that branch failed with "attempt to select less than one element". This also covers predict(type = "raw", opts = ...) (#1408).
set_model_arg() now stores func as a list when given the named character vector form shown in its documentation, such as c(pkg = "dials", fun = "mixture"). That form registered without complaint but made tuning fail much later with $ operator is invalid for atomic vectors (#1251).
svm_linear() with the "LiblineaR" engine now passes cost to LiblineaR::LiblineaR(). It was previously mapped to a nonexistent C argument, which the engine silently absorbed into its dots, so every fit used the engine default of cost = 1 and tuning over cost had no effect. Fitted results will change for any model with a non-default cost (#1405).
Fitting with sparse data now respects the model mode, so loading an extension package that registers an engine for a different mode can no longer alter sparse data support for the original mode (#1382).
Corrected documentation that referred to fit() and fit_xy() as arguments rather than functions in the case weights template (#1394).

From NEWS.md in the release

Install

R (CRAN)pak (dev)Ubuntu (r2u)Fedora (COPR)openSUSE (OBS)

install.packages("parsnip", repos = "https://cloud.r-project.org/")

Installs the current CRAN release. Prebuilt binaries on Windows and macOS.

if (!requireNamespace("pak", quietly = TRUE))
  install.packages("pak", repos = "https://cloud.r-project.org/")
pak::pak("tidymodels/parsnip")

Development version from tidymodels/parsnip. Installs pak first if it is not already available.

Release

# enable cranapt (r2u) if it is not already set up, then install
grep -rqs r2u.stat.illinois.edu /etc/apt/sources.list /etc/apt/sources.list.d/ \
  || curl -fsSL https://raw.githubusercontent.com/eddelbuettel/r2u/master/inst/scripts/add_cranapt_noble.sh | sudo bash
sudo apt install -y r-cran-parsnip

Fast Ubuntu binaries via r2u (no compilation). The cranapt setup runs once and is skipped when it is already enabled. We list this channel from recorded r2u download activity rather than from a live index of the repository, so treat it as last known rather than guaranteed.

bspm (used by r2u) also needs make: sudo apt install make

# enable the iucar/cran COPR if it is not already present, then install
ls /etc/yum.repos.d/*iucar*cran*.repo >/dev/null 2>&1 \
  || sudo dnf copr enable -y iucar/cran
sudo dnf install -y R-CRAN-parsnip

RPMs from the iucar/cran COPR, rebuilt from CRAN.

For install.packages() integration in R, also: sudo dnf install R-CoprManager

# add the autoCRAN OBS repo (Tumbleweed) if it is missing, then install
zypper -q lr | grep -qi autoCRAN \
  || sudo zypper addrepo https://download.opensuse.org/repositories/devel:/languages:/R:/autoCRAN/openSUSE_Tumbleweed/devel:languages:R:autoCRAN.repo
sudo zypper install -y R-parsnip

RPMs auto-built from CRAN on openSUSE OBS (Tumbleweed). We list this channel from recorded autoCRAN download activity rather than from a live index of the repository, so treat it as last known rather than guaranteed.

Health

CRAN checks

13OK

CRAN check page ↗

Slowest check: 7.5 min · r-oldrel-windows-x86_64

Status changes

OK13 OK2026-10-04
ERROR8 OK, 4 NOTE, 1 ERROR2026-09-11
NOTE12 OK, 1 NOTE2026-07-11
OK13 OK2026-06-09
ERROR12 OK, 1 ERROR2026-06-08

Show 3 earlier changesHide 3 earlier changes

OK14 OK2026-04-10
ERROR13 OK, 1 ERROR2026-04-07
OK14 OK2026-03-10

Advisories

none known

How this is measured·How advisories are found

Documentation

In release 1.6.1

READMEYes · 410 wordsVignettesYes · 1 · dynamicNEWSYes · 100% structured · mentions 1.6.1Citation fileNoWebsiteYes · pkgdown

The repository also keeps web-only articles.

Help pages: 242 (173 internal)includes a package overview page
Examples that run: 100%58 run in a plain check (57 of them under a condition)
Exported functions with examples: 48%
Documented parameters: 96%
Return-value docs: 39%
Help pages citing references: 18%

Vignettes (1)

Introduction to parsnip HTML

How these are measured

Downloads

506.7K

CRAN downloads in the past year

Rank #591 · ~1.4K/day · ~42.2K/mo

Releases R versions

Daily download trend is not available in this view yet.

35.7K30 days

105.4K90 days

506.7K1 year

Compare downloads with other packages →

Also on14.7K r2u9 autocran39K conda_forge218 c2d4u

How downloads are counted

Repository

tidymodels/parsnip ↗Organization: tidymodels

661Stars

116Forks

66Open issues

3Open PRs

36Releases

2,227Commits

67Commit identities

2,227 commits · Last activity 2026-09-27 · +0.3% stars, 30d · Default branch at 1.6.1.9000, ahead of CRAN 1.6.1 (checked 2026-10-11)

Responsiveness

How these are sampled

PRs merged (all time): 96%
Median days to close, 50 most recently updated closed issues: 120
Median days to close, 50 most recently updated closed PRs: 0

Stars over time

2025-09-29 · 6312026-10-06 · 661

How these are measured

PracticesBeta

14 development-tooling and community-health practices detected across 5 families in the upstream repository

Checks run against github.com/tidymodels/parsnip on 2026-10-11.

Continuous integration (2)

GitHub ActionsR CMD check workflow

The R CMD check workflow runs on Linux, macOS and Windows, and on R-devel

Governance and community (3)

Code of conductContributing guideIssue templates

Docs source (4)

README.Rmdpkgdown

GitHub Pages site last deployed 2026-09-27 · The repository has a site deploy workflow

pkgdown site last built 2026-09-27

Coverage (1)

Coverage reporting (Codecov)

Show all practicesHide 1 practice

Lint, format, editor (4)

Lint workflowAirVS CodeRStudio project

How this is detected·Detection ruleset v3 (2026-09-26)

AI coding toolsBeta

This repository shows signs of AI coding tools.

Claude Codefirst seen 2026-03-02
Seen in the repository: .claude, .claude/skills
Named in .Rbuildignore: .claude
Commits: at least 1 crediting it in the message
GitHub Copilotfirst seen 2026-07-27
Pull request: opened by its own account
Commits: 1 by its own account · at least 1 crediting it in the message

Shared conventions any agent may read, so they name no particular tool.

Seen in the repository: AGENTS.md

From outside contributors

Claude Code
Pull requests: 1 from outside contributors, first on 2026-09-17, 1 from forks

Models named in its commits

Claude Fable 5.11 commit · Sep 2026

Counts are commits, not lines of code. Commits by a tool's accounts count once across all of its accounts. Commits crediting it are the largest count among its credit lines, a minimum.How this is detected·Detection ruleset v4 (updated 2026-09-27)

Dependencies

Declared dependencies

39 external dependencies (excludes base and recommended)

Installing parsnip adds 33 packages (16 with compiled code, 29 MB of source tarballs). Heaviest: stringi 11 MB, ggplot2 6.0 MB, isoband 1.5 MB.

Depends (1)

R >= 4.1

Imports (19)

cli dplyr generics ggplot2 globals glue hardhat lifecycle magrittr pillar prettyunits purrr rlang sparsevctrsstatstibble tidyrutilsvctrs

LinkingTo (0)

none

Suggests (29)

bench C50 covr dials earth ggrepel keras keras3 kernlab kknn knitr LiblineaRMASSMatrixmethodsmgcvmodeldatanlmeprodlim ranger remotes rmarkdownrpartsparklyrsurvivaltensorflow testthat withr xgboost

Enhances (0)

none

Dependency license breakdown →

Reverse dependencies

58direct

19indirect

Depends 15 · Imports 43

Suggested by 50

Show the 48 most downloaded dependentsHide the 48 most downloaded dependents

Depends, Imports or LinkingTo, most downloaded first

and 10 more dependents and 50 packages that suggest it on CRAN ↗

How these are counted

Code & Tests

Datasets

People & History

People (4)

Max Kuhn

Maintainer since 0.0.1 (2018), Author

Davis Vaughan

Author

Emil Hvitfeldt

Contributor · added in 0.1.6

Posit Software, PBC

Copyright holder, Funder · added in 1.1.0

Listed in earlier versions (2)

RStudio

no longer listed · 0.0.1 to 1.0.3

Posit Software PBC

no longer listed · 1.0.4

Releases (34)

Pick two releases to compare their code metrics.

1.6.1Latest
record
2026-09-27 · current release · diff ↗ ·
added 8 exports
check_outcome_levels, tabular_auto_int, tabular_chronos, tabular_icl, tabular_pfn, tabular_resnet, tabular_rln, tabular_saint
1.6.0
record
2026-05-14 · diff ↗ ·
added 3 exports
keras3_activations, keras3_mlp, keras3_predict_classes
R
R 4.6.0 released · 2026-04-24
1.5.0
record
2026-04-09 · diff ↗ ·
added 8 exports
format_predictions, mcqrnn_train, model_arg_exists, ordinal_reg, predict_class, predict_classprob, predict_hazard, predict_quantile
1.4.1
record
2026-01-11 · diff ↗
1.4.0
record
2025-12-01 · diff ↗
1.3.3
record
2025-08-31 · diff ↗
1.3.2
record
2025-05-28 · diff ↗
R
R 4.5.0 released · 2025-04-11
1.3.1
record
2025-03-12 · diff ↗
1.3.0
record
2025-02-14 · diff ↗ ·
removed 2 exports
bartMachine_interval_calc, rpart_train

added 5 exports
.get_prediction_column_names, ensure_parsnip_format, extract_fit_time, keras_activations, matrix_to_quantile_pred
R
R 4.4.0 released · 2024-04-24
1.2.1
record
2024-03-22 · diff ↗
1.2.0
record
2024-02-16 · diff ↗ ·
added 5 exports
case_weights_allowed, check_args, predict_confint, predict_predint, predict_predint.model_fit
1.1.1
record
2023-08-17 · diff ↗ ·
added 2 exports
.extract_surv_status, .extract_surv_time
R
R 4.3.0 released · 2023-04-21

Show 27 earlier eventsHide 27 earlier events

1.1.0
record
2023-04-12 · diff ↗ ·
added 1 export
.censoring_weights_graf
1.0.4
record
2023-02-23 · diff ↗ ·
removed 2 exports
check_model_doesnt_exist, check_model_exists
1.0.3
record
2022-11-11 · diff ↗ ·
added 3 exports
prompt_missing_implementation, spec_is_loaded, spec_is_possible
1.0.2
record
2022-10-01 · diff ↗ ·
added 2 exports
bag_mlp, condense_control
1.0.1
record
2022-08-18 · diff ↗ ·
added 1 export
auto_ml
1.0.0
record
2022-06-16 · diff ↗ ·
added 17 exports
.model_param_name_key, cforest_train, ctree_train, format_class, format_classprobs, format_hazard, format_linear_pred, format_num, format_survival, format_time, frequency_weights, glm_grouped, importance_weights, max_mtry_formula, print_model_spec, update_spec, xgb_predict
R
R 4.2.0 released · 2022-04-22
0.2.1
record
2022-03-17 · diff ↗
0.2.0
record
2022-03-09 · diff ↗ ·
added 26 exports
C5_rules, autoplot, bag_mars, bag_tree, bart, bartMachine_interval_calc, check_model_doesnt_exist, check_model_exists, cubist_rules, dbart_predict_calc, discrim_flexible, discrim_linear, discrim_quad, discrim_regularized, extract_parameter_dials, extract_parameter_set_dials, keras_predict_classes, knit_engine_docs, list_md_problems, naive_Bayes, pls, poisson_reg, rule_fit, set_tf_seed, tune, update_model_info_file
0.1.7
record
2021-07-21 · diff ↗ ·
added 13 exports
.check_glmnet_penalty_fit, .check_glmnet_penalty_predict, .convert_form_to_xy_fit, .convert_form_to_xy_new, .convert_xy_to_form_fit, .convert_xy_to_form_new, .organize_glmnet_pred, extract_fit_engine, extract_spec_parsnip, find_engine_files, gen_additive_mod, make_engine_list, make_seealso_list
0.1.6
record
2021-05-27 · diff ↗ ·
added 11 exports
predict_hazard.model_fit, predict_linear_pred, predict_linear_pred.model_fit, predict_survival, predict_survival.model_fit, predict_time, predict_time.model_fit, proportional_hazards, required_pkgs, survival_reg, svm_linear
R
R 4.1.0 released · 2021-05-18
0.1.5
record
2021-01-19 · diff ↗ ·
added 3 exports
augment, parsnip_addin, stan_conf_int
0.1.4
record
2020-10-27 · diff ↗ ·
added 7 exports
make_call, maybe_data_frame, maybe_matrix, min_cols, min_rows, prepare_data, show_engines
0.1.3
record
2020-08-04 · diff ↗ ·
added 2 exports
glance, update_engine_parameters
0.1.2
record
2020-07-03 · diff ↗ ·
removed 1 export
convert_args

added 5 exports
contr_one_hot, get_encoding, repair_call, req_pkgs, set_encoding
0.1.1
record
2020-05-06 · diff ↗ ·
removed 1 export
tidy.model_fit

added 2 exports
predict_raw, tidy
R
R 4.0.0 released · 2020-04-24
0.1.0
record
2020-04-09 · diff ↗ ·
added 11 exports
convert_args, convert_stan_interval, eval_args, predict_class.model_fit, predict_classprob.model_fit, predict_confint.model_fit, predict_numeric, predict_numeric.model_fit, predict_quantile.model_fit, predict_raw.model_fit, translate.default
0.0.5
record
2020-01-07 · diff ↗ ·
added 7 exports
check_final_param, is_varying, new_model_spec, null_value, show_fit, update_dot_check, update_main_parameters
0.0.4
record
2019-11-02 · diff ↗ ·
removed 7 exports
min_grid, min_grid.boost_tree, min_grid.linear_reg, min_grid.logistic_reg, min_grid.mars, min_grid.multinom_reg, min_grid.nearest_neighbor

added 1 export
control_parsnip
0.0.3.1
record
2019-08-06 · diff ↗
0.0.3
record
2019-07-31 · diff ↗ ·
added 26 exports
add_rowindex, get_dependency, get_fit, get_from_env, get_model_env, get_pred_type, has_multi_predict, min_grid, min_grid.boost_tree, min_grid.linear_reg, min_grid.logistic_reg, min_grid.mars, min_grid.multinom_reg, min_grid.nearest_neighbor, multi_predict_args, pred_value_template, set_dependency, set_env_val, set_fit, set_in_env, set_model_arg, set_model_engine, set_model_mode, set_new_model, set_pred, show_model_info
R
R 3.6.0 released · 2019-04-26
0.0.2
record
2019-03-22 · diff ↗ ·
removed 17 exports
predict_class, predict_class.model_fit, predict_classprob, predict_classprob.model_fit, predict_confint, predict_confint.model_fit, predict_numeric, predict_numeric.model_fit, predict_predint, predict_predint.model_fit, predict_quantile, predict_quantile.model_fit, predict_raw, predict_raw.model_fit, varying_args.model_spec, varying_args.recipe, varying_args.step

added 3 exports
null_model, nullmodel, tidy.model_fit
0.0.1
record
2018-11-12
R
R 3.5.0 released · 2018-04-23

Package metadata

First published: 2018-11-12
Total releases: 34 / 8 yrs
License: MIT + file LICENSE OSI · relicensed from GPL-2 in 0.1.6
Minimum R: ≥ 4.1
Bundled data: 3.6 KB / 1 file
Download size: 864 KB

Cite

Cite this package

Run in R for the authors' preferred citation:

citation("parsnip")

Format

Kuhn, M., Posit Software, PBC, Hvitfeldt, E., & Vaughan, D. (2026). parsnip: A Common API to Modeling and Analysis Functions (Version 1.6.1) [Computer software]. https://doi.org/10.32614/CRAN.package.parsnip

This package has no citation file, so citation() returns this entry built from DESCRIPTION.

Cite the R Observatory

For a number measured here: a download total, a coverage figure, an archival date.

Format

APA

Balamuta, J. J. (2026). R Observatory: Metrics for parsnip version 1.6.1 [Data set]. HJJB, LLC. Data release v2026-10-17. https://doi.org/10.5281/zenodo.21843040

From data release v2026-10-17, which the citation names so these numbers can be found later. More on citing and the projects behind them.

Report a problem with this page →