Get the Insights of Your Neural Network
Overview
Interpretation methods for analyzing the behavior and individual predictions of modern neural networks in a three-step procedure: Converting the model, running the interpretation method, and visualizing the results. Implemented methods are, e.g., 'Connection Weights' described by Olden et al. (2004) doi:10.1016/j.ecolmodel.2004.03.013, layer-wise relevance propagation ('LRP') described by Bach et al. (2015) doi:10.1371/journal.pone.0130140, deep learning important features ('DeepLIFT') described by Shrikumar et al. (2017) doi:10.48550/arXiv.1704.02685 and gradient-based methods like 'SmoothGrad' described by Smilkov et al. (2017) doi:10.48550/arXiv.1706.03825, 'Gradient x Input' or 'Vanilla Gradient'. Details can be found in the accompanying scientific paper: Koenen & Wright (2024, Journal of Statistical Software, doi:10.18637/jss.v111.i08).
- Fixed a bug in the
evalargument in vignette chunks. - Skipped some tests for convolutional layers on Windows due to issues with
functional convolutions when using non-default dilation and double precision
tensors (see PyTorch issue #141221).
From NEWS.md in the release
Install
Health
NOTERd cross-referencesr-devel-linux-x86_64-debian-clang, r-devel-linux-x86_64-debian-gcc, r-devel-linux-x86_64-fedora-clang, r-devel-linux-x86_64-fedora-gcc, r-patched-linux-x86_64, r-release-linux-x86_64
Unknown package ‘fastshap’ in Rd xrefs
- NOTE7 OK, 6 NOTE2026-08-04
- ERROR6 OK, 6 NOTE, 1 ERROR2026-07-30
- NOTE7 OK, 6 NOTE2026-06-09
- ERROR7 OK, 5 NOTE, 1 ERROR2026-06-08
- NOTE11 OK, 2 NOTE2026-05-28
Show 1 earlier changeHide 1 earlier change
- OK14 OK2026-03-10
Documentation
The repository also keeps web-only articles.
- Help pages
- 26includes a package overview page
- Examples that run
- 100%11 run in a plain check (11 of them under a condition)
- Exported functions with examples
- 73%
- Documented parameters
- 93%
- Return-value docs
- 7%
- Help pages citing references
- 35%
Downloads
Repository
- PRs merged (all time)
- 87%
- Median days to close, 50 most recently updated closed issues
- 243
- Median days to close, 50 most recently updated closed PRs
- 0
9 development-tooling and community-health practices detected across 5 families in the upstream repository
Checks run against github.com/bips-hb/innsight on 2026-10-11.
The R CMD check workflow runs on Linux, macOS and Windows, and on R-devel
GitHub Pages site last deployed 2026-02-22 · The repository has a site deploy workflow
pkgdown site last built 2026-02-22
Show all practicesHide 1 practice
Dependencies
Installing innsight adds 32 packages (17 with compiled code, 20 MB of source tarballs). Heaviest: ggplot2 6.0 MB, torch 1.9 MB, Rcpp 1.8 MB.
Code & Tests
People & History
Pick two releases to compare their code metrics.
- RR 4.6.0 released · 2026-04-24
- RR 4.5.0 released · 2025-04-11
- 0.3.2Latest
- 0.3.12024-11-26 · diff ↗
- RR 4.4.0 released · 2024-04-24
- 0.3.02023-12-21 · diff ↗ ·
added 18 exports
AgnosticWrapper, DeepSHAP, ExpectedGradient, IntegratedGradient, LIME, SHAP, convert, plot_global, run_cw, run_deeplift, run_deepshap, run_expgrad, run_grad, run_intgrad, run_lime, run_lrp, run_shap, run_smoothgrad - RR 4.3.0 released · 2023-04-21
- 0.2.0
- 0.1.12022-08-29 · diff ↗
- RR 4.2.0 released · 2022-04-22
- 0.1.02021-11-22
- RR 4.1.0 released · 2021-05-18
Package metadata
- First published
- 2021-11-22
- Total releases
- 6 / 5 yrs
- License
- MIT + file LICENSE OSI
- Minimum R
- ≥ 3.5.0
- Download size
- 1.9 MB
Cite
Cite this package
Run in R for the authors' entry:
citation("innsight")The package provides a citation file, so citation() returns the authors' own entry.
It points to 10.18637/jss.v111.i08.
Cite the R Observatory
For a number measured here: a download total, a coverage figure, an archival date.
From data release v2026-10-18, which the citation names so these numbers can be found later. More on citing and the projects behind them.