rMVPA is an R library for multivariate pattern analysis of neuroimaging data. The goal of this library is to make MVPA analyses easy. It can be used both programmatically from within R or using a command line interface. ‘rMVPA’ provides a lightweight model registry and efficient resampling methods for machine learning. What rMVPA provides is the infrastructure for conducting machine learning analyses on neuroimaging data.

Documentation and vignettes: https://bbuchsbaum.github.io/rMVPA/

Method note: ITEM vs continuous-time hrfdecoder paths: inst/notes/item_vs_hrfdecoder.md

Searchlight execution now uses a deterministic optimized runtime policy by default (no ambient option/profile/env switches):

• fold-cache reuse
• geometry cache reuse
• clustered-neighbor fastpath
• matrix-first ROI processing
• fast ROI filtering
• RSA fast kernel
• naive cross-decoding fast kernel
• backend policy default = auto
• SWIFT multiclass searchlight engine (eligible multiclass, standard CV paths; outputs Accuracy, AUC, and SWIFT_Info)

To force a specific backend for a call, pass backend= explicitly:

run_searchlight(mspec, radius = 8, method = "standard", backend = "default")

To force a specific searchlight engine for a call, pass engine= explicitly:

run_searchlight(mspec, radius = 8, method = "standard", engine = "legacy")

Engine selection is recorded on results for auditability:

• attr(res, "searchlight_engine")
• attr(res, "searchlight_engine_requested")
• attr(res, "searchlight_engine_resolved")
• attr(res, "searchlight_engine_reason")
• attr(res, "searchlight_engine_fallback_error") (only when fallback occurred)

Operational rollback and release checklist: docs/searchlight_performance_rollbacks.md

Perf guardrail harness (scripts/run_perf_guardrails.R) now enforces a singleton lock by default to prevent concurrent local runs from polluting timings. Override controls:

• RMVPA_PERF_SINGLETON=true|false (default true)
• RMVPA_PERF_LOCK_DIR=/path/to/lock (default ./.tmp/perf_guardrails.lock)
• RMVPA_PERF_LOCK_STALE_SECONDS=<seconds> (default 21600)

To install rMVPA from within R, use the devtools function install_github. You will need the development version of neuroim2 as well.

From within R:

#library(devtools)
install_github("bbuchsbaum/neuroim2")
install_github("bbuchsbaum/rMVPA")

git clone git@github.com:bbuchsbaum/rMVPA.git
R CMD install rMVPA

wget https://raw.githubusercontent.com/bbuchsbaum/rMVPA/master/scripts/MVPA_Searchlight.R

wget https://raw.githubusercontent.com/bbuchsbaum/rMVPA/master/scripts/MVPA_Regional.R

Then, move these files to a folder on your PATH and make them executable:

chmod +x MVPA_Searchlight.R

chmod +x MVPA_Regional.R

This package uses the albersdown theme. Vignettes are styled with vignettes/albers.css and a local vignettes/albers.js; the palette family is provided via params$family (default 'red'). The pkgdown site uses template: { package: albersdown }.

This package uses the albersdown theme. Vignettes are styled with vignettes/albers.css and a local vignettes/albers.js; the palette family is provided via params$family (default 'red'). The pkgdown site uses template: { package: albersdown }.

This package uses the albersdown theme. Existing vignette theme hooks are replaced so albers.css and local albers.js render consistently on CRAN and GitHub Pages. The palette family is provided via params$family (default 'red'). The pkgdown site uses template: { package: albersdown }.