// Add new computational engines to existing parsnip models. Use when connecting an existing parsnip model (linear_reg, boost_tree, etc.) to a new computational backend or R package.
| name | add-parsnip-engine |
| description | Add new computational engines to existing parsnip models. Use when connecting an existing parsnip model (linear_reg, boost_tree, etc.) to a new computational backend or R package. |
Guide for adding new engines to existing parsnip models. This skill covers
registering engines (like adding "spark" to linear_reg()) without creating
entirely new model types.
Use this skill when: Adding a new engine to an existing parsnip model type.
For creating new models: See add-parsnip-model skill instead.
This skill supports two distinct development contexts:
Creating a new R package that adds engines to existing parsnip models.
✅ Use this for: New packages, standalone engines, CRAN submissions
📦 Package detection: No parsnip in DESCRIPTION's Package: field
⚠️ Constraint: Can only use exported functions (no :::)
📖 Guide: Extension Development Guide
Contributing directly to parsnip via pull requests.
✅ Use this for: Contributing to tidymodels/parsnip repository
📦 Package detection: Package: parsnip in DESCRIPTION
✨ Benefit: Can use internal functions and package infrastructure
📖 Guide: Source Development Guide
This main guide shows extension development patterns. If you're contributing to parsnip itself, see the Source Development Guide for package-specific details.
INSTRUCTIONS FOR CLAUDE: Run the verification script first to determine the development context:
Rscript -e 'source(Sys.glob(path.expand("~/.claude/plugins/cache/tidymodels-skills/tidymodels-dev/*/tidymodels/shared-references/scripts/verify-setup.R"))[1])'
Then follow the appropriate path based on the output:
Output: "All checks for source development complete." → Go to Source Development Guide
Output: "All checks for extension development complete." (no warnings) → Go to Extension Development Guide
Output: Shows "Warning - [UUID]" messages → Go to Extension Prerequisites to resolve warnings first
Adding an engine to an existing parsnip model provides:
Connection to new computational backends (e.g., H2O, Spark, TensorFlow)
Standardized interface with parsnip models
Support for multiple prediction types
Integration with tidymodels ecosystem
Consistent API regardless of engine
What this skill covers:
Planning and choosing the right interface
Complete registration sequence
Fit and predict method implementation
Testing engine implementations
Multi-mode support (regression + classification)
INSTRUCTIONS FOR CLAUDE: Check if repos/parsnip/ exists in the current
working directory. Use this to guide development:
If repos/parsnip/ exists:
✅ Use it as a reference throughout development
Read source files (e.g., repos/parsnip/R/linear_reg_data.R) to study engine
registration patterns
Read test files (e.g., repos/parsnip/tests/testthat/test-linear_reg.R) for
testing patterns
Reference these files when answering complex questions or solving problems
Look at actual code structure, validation patterns, and edge case handling
If repos/parsnip/ does NOT exist:
Suggest cloning the repository using the scripts in Repository Access Guide
This is optional but strongly recommended for high-quality development
If the user declines, reference files using GitHub URLs:
Format: https://github.com/tidymodels/parsnip/blob/main/R/[file-name].R
Example: https://github.com/tidymodels/parsnip/blob/main/R/linear_reg_data.R
This allows users to click through to see implementations
When to use repository references:
Complex implementation questions (e.g., "How does parsnip handle multi-mode engines?")
Debugging issues (compare user's code to working implementation)
Understanding patterns (study similar engines)
Test design (see how parsnip tests edge cases)
Architecture decisions (understand internal structure)
See Repository Access Guide for setup instructions.
Development Guides:
Extension Development Guide - Creating new packages that add engines
Source Development Guide - Contributing PRs to parsnip itself
Core Implementation References:
Engine Implementation - Complete registration sequence, examples, patterns
Fit and Predict Methods - Implementation details for fit/predict
Prediction Types - All 11 prediction types
Mode Handling - Multi-mode support (regression
Encoding Options - Interface types and data conversion
Model-Specific Guides:
Shared References (Extension Development):
Extension Prerequisites - Package setup
Development Workflow - Fast iteration cycle
Extension Requirements - Complete guide:
Source Development Specific:
⚠️ IMPORTANT: Before implementing engines, complete the extension prerequisites sequence:
👉 Extension Prerequisites Guide
This guide includes critical steps like use_claude_code() (if available) that
must run BEFORE adding dependencies. Following the complete sequence ensures
proper package initialization and Claude Code integration.
After completing extension prerequisites, return here to implement your engine.
Parsnip Fundamentals:
Before adding an engine, understand:
How parsnip models work - Model Specification System
Fit and predict patterns - Fit and Predict Methods
Available output formats - Prediction Types
INSTRUCTIONS FOR CLAUDE: Assess complexity first, then choose approach:
Single mode (regression OR classification, not both)
Formula interface OR matrix interface (pick one)
1-3 parameters to map
Standard prediction type (numeric OR class/prob)
→ Use streamlined approach:
Target 2 files: R/zzz.R (15-30 lines), tests/testthat/test-*.R; acceptable to reach 4-6 if needed
NO summary docs, NO example files
Multi-mode (regression AND classification)
Matrix interface with encoding
Survival/censored regression
Custom prediction post-processing
→ Reference detailed guides:
See Mode Handling for multi-mode
See Encoding Options for matrix interfaces
Still target 2-3 files (R/zzz.R, tests, optional README); acceptable to reach 4-6 if implementation requires it
Core registration steps:
set_model_engine()set_dependency()set_model_arg()set_fit()set_encoding() (if needed)set_pred()File Discipline:
Extension: Create 2-3 files (R/zzz.R, tests/testthat/test-*.R, optional README.md); acceptable to reach 4-6 files if implementation requires it
Source: Modify 1-2 files (add to R/_data.R, add to tests/testthat/test-.R); acceptable to reach 3-7 files if implementation requires it
Never create: IMPLEMENTATION_SUMMARY.md, example_usage.R, helper files
See Engine Implementation Guide for complete details and examples.
The registration process differs slightly by context:
Extension Development:
Register in .onLoad() function
Use parsnip:: prefix for all functions
Cannot access internal helpers
Create function that contains all registrations
Source Development:
Add to existing R/[model]_data.R file
No prefix needed for parsnip functions
Can use internal helpers if needed
Follow existing file organization patterns
See respective guides for detailed registration patterns.
Essential tests to include:
Engine fits successfully
Formula and xy interfaces work (if applicable)
Each prediction type returns correct format
Predictions match data dimensions
Factor handling works correctly
Error messages are clear
See testing guides:
Extension: Testing Patterns (Extension)
Source: Testing Patterns (Source)
Add an engine when:
Model type already exists in parsnip
Engine provides different computational approach
Engine offers performance benefits or unique features
Package is well-maintained and stable
Don't add an engine when:
Model type doesn't exist (see add-parsnip-model instead)
Engine is functionally identical to existing
Package is experimental or unmaintained
Only cosmetic differences from existing engines
add-parsnip-model - Create new model specifications (if model doesn't exist yet)
add-dials-parameter - Define tunable parameters for engine arguments
add-recipe-step - Preprocess data before model fitting
add-yardstick-metric - Evaluate engine predictions with custom metrics
For Extension Development (creating new packages):
For Source Development (contributing to parsnip):
R/[model]_data.R fileFor questions or contributions, see:
Create a new preprocessing step for the recipes package following tidymodels conventions
Guide for creating new yardstick metrics. Use when a developer needs to extend yardstick with a custom performance metric, including numeric, class, probability, ordered probability, survival (static, dynamic, integrated, linear predictor), and quantile metrics.
Build machine learning models using tidymodels for tabular data using proper data spending, resampling, and validation practices. Covers train/test splitting, cross-validation, feature engineering, model tuning, and evaluation. Use when building predictive models, comparing algorithms, or when users mention machine learning, model training, or prediction tasks.