RMVMR
RMVMR is being tidied in lockstep with MVMR, the package it wraps
Comparison · Analytics
kernelshap vs units
A side-by-side editorial comparison of kernelshap and units — release velocity, themes, recent moves, and the top alternatives to consider.
Shared themes:r package
kernelshap vs units: at a glance
What is kernelshap?
kernelshap makes permutation SHAP practical past eight features, then fixes the kernel weights it had wrong.
kernelshap computes model-agnostic SHAP values in R through Kernel SHAP, permutation SHAP and an exact additive explainer. Version 0.8.0 added a sampling permutation-SHAP algorithm with standard errors and early stopping, lifting the practical feature ceiling past what the exact method allows. Version 0.9.0 then corrected a bug in how kernel weights were computed — exact Kernel SHAP now agrees with exact permutation SHAP — and moved parallelism from foreach to doFuture.
What is units?
units reaches 1.0 by rewriting its parser and accepting the breakage that comes with it.
units attaches physical units to R vectors on top of the udunits2 library, covering arithmetic, conversion and ggplot2 scales. Version 1.0-0 replaced the unit-expression tokenizer so numbers are consistently treated as prefixes, and expressions like ml/min/1.73m^2 now parse the way physiologists write them. Printing follows NIST conventions, and 1.0-1 is a fix pass over memory handling and parser edge cases.
kernelshap vs units: editorial side-by-side
kernelshap
0.0ANALYTICS
kernelshap makes permutation SHAP practical past eight features, then fixes the kernel weights it had wrong.
◆ Current state
kernelshap computes model-agnostic SHAP values in R through Kernel SHAP, permutation SHAP and an exact additive explainer. Version 0.8.0 added a sampling permutation-SHAP algorithm with standard errors and early stopping, lifting the practical feature ceiling past what the exact method allows. Version 0.9.0 then corrected a bug in how kernel weights were computed — exact Kernel SHAP now agrees with exact permutation SHAP — and moved parallelism from foreach to doFuture.
◆ Where it's heading
Two concerns drive this package: making exact methods reach further, and being demonstrably right. The first shows in the additive explainer, the optional background dataset and the sampling permutation algorithm; the second in unit tests written against Python's shap, credited fixes from outside contributors, and a willingness to ship a correctness fix that changes numbers people have already published. Speed work runs continuously underneath — direct solves replacing the Moore-Penrose pseudo-inverse, roughly 10% less memory.
◆ Prediction
The 0.6.0 and 0.7.0 notes each promised a stable 1.0.0 that has not arrived; with the weighting bug fixed and parallelism reworked, a 1.0 release is the most plausible next step.
units
0.0ANALYTICS
units reaches 1.0 by rewriting its parser and accepting the breakage that comes with it.
◆ Current state
units attaches physical units to R vectors on top of the udunits2 library, covering arithmetic, conversion and ggplot2 scales. Version 1.0-0 replaced the unit-expression tokenizer so numbers are consistently treated as prefixes, and expressions like ml/min/1.73m^2 now parse the way physiologists write them. Printing follows NIST conventions, and 1.0-1 is a fix pass over memory handling and parser edge cases.
◆ Where it's heading
The long 0.8-x run was accretion — ggplot2 scales absorbed from ggforce, ud_convert(), matrix methods, steady performance work — while known parsing defects stayed in place. The 1.0 release finally traded backwards compatibility for correct parsing, and 1.0-1's pointer-wrapping and exception-propagation work suggests the C++ glue is being hardened behind it. Fixes cluster at the udunits2 boundary, which remains the main source of surprises.
◆ Prediction
Expect continued patch releases against udunits2 quirks and the new tokenizer's fallout rather than new surface area, with the ggplot2 integration and conversion helpers already in place.
Alternatives to kernelshap and units
Other Analytics products tracked by Sparkpulse, ranked by recent ship velocity. Each card links to a full editorial trajectory and lets you pivot into a head-to-head comparison with either kernelshap or units.
geoarrow-r
geoarrow tracks the GeoArrow spec and otherwise just keeps compiling
n2khab
n2khab keeps retracting interpretations of habitat data it can't actually support
tidypolars
tidypolars is grinding toward complete dplyr coverage, one supported function at a time
OneSampleMR
OneSampleMR found that argument order in a formula was silently changing its estimates
bpbounds
bpbounds found the same swapped-cell bug twice and clamped its bounds back into range
See all kernelshap alternatives → · See all units alternatives →
Recent activity from kernelshap and units
Latest ship moves from both products, interleaved chronologically. ⚡ = editorial spark.
- 5mo agounitsMemory-leak and udunits2 parsing fixes after 1.0
- 10mo agounitsNew tokenizer parses compound units correctly ⚡
- 1y agokernelshapKernel weight bug fixed; parallelism moves to doFuture
- 1y agokernelshapSampling permutation SHAP with standard errors ⚡
- 1y agounitsCopy semantics fix in ud_convert(); C++17 for old R
- 1y agounitscbind/rbind methods, ud_convert() and broad fixes
- 1y agounitsSilences a CRAN compiler warning
- 1y agokernelshapBackground data now optional; ranger survival support
- 2y agokernelshapFactor-valued predictions dropped
- 2y agokernelshapadditive_shap() explains additive models exactly
- 2y agokernelshapFaster on plain data.frames
- 2y agounitsRestores simplify=FALSE for identical units
Frequently asked questions
What is the difference between kernelshap and units?
Both compete on the same themes — r package — within Analytics. kernelshap and units are shipping at a similar cadence (velocity 0.0 vs 0.0, both within Sparkpulse's "active" band). See the at-a-glance table above for a side-by-side breakdown of velocity, recent sparks, and editorial themes.
Is kernelshap better than units?
Sparkpulse doesn't pick a winner — we score release velocity, not feature parity. kernelshap and units are shipping at a similar cadence (velocity 0.0 vs 0.0, both within Sparkpulse's "active" band). For your specific use case, the alternatives sections above list other Analytics products to evaluate alongside.
What are the best alternatives to kernelshap?
Top kernelshap alternatives in Analytics are ranked by recent ship velocity. Browse the "kernelshap alternatives" section above for the current picks, or visit /alternatives/kernelshap for the full list with editorial commentary on each.
What are the best alternatives to units?
Top units alternatives in Analytics are ranked by recent ship velocity. Browse the "units alternatives" section above for the current picks, or visit /alternatives/units-r for the full list with editorial commentary on each.