LoginSign Up
← Back to all sparks
S

spEDM

ANALYTICS
Velocity0.0

R package spEDM by stscl — release notes from GitHub.

github.com/stscl/spEDM ↗

Spatial causal discovery in R, one exposed method per release

causal-inferencespatial-analysisempirical-dynamic-modelingr-packagebreaking-changescpp-backend
◆Current state
spEDM brings empirical dynamic modeling to spatial data — cross mapping, convergent cross mapping and pattern causality over spatial vector and raster inputs, with the numerics in C++ behind S4 generics. The recent releases have exposed geographical pattern causality and spatially convergent partial cross mapping at the R level with vignettes, and 1.12 turns to consolidating the API. It is part of the stscl family alongside the temporal-domain tEDM, with which it shares both its C++ core and its maintainer.
◆Where it's heading
The cadence is steady and predictable: each release surfaces one more EDM method as an R-level API with a vignette, then spends the rest of its notes on parameter-handling consistency across the generics. Breaking changes are frequent and deliberate — argument renames, parameter reordering, NA-handling defaults — which reads as a package still settling its interface while the method surface expands. Shared changes appear in tEDM within days, so interface churn lands on both packages at once.
◆Prediction
Expect the next release to expose another causality variant at the R level with an accompanying vignette, and to continue renaming or reordering parameters toward consistency across the spatial and temporal packages.

◆Recent moves

  1. 4mo ago

    Data slicing for large-scale pattern causality, plus API breaks

    Adds library/prediction-set slicing so large-scale pattern causality computations become tractable, and allows asymmetric (E, tau, k) configurations in bidirectional analysis. The breaking half is larger: causality detection direction is realigned with variable naming, k is constrained to be at least E, NA_rm becomes na_rm and defaults to true, and conds in scpcm now takes condition variables in direct path order. Callers should expect to edit code.

    View source ↗
  2. 6mo ago

    spEDM 1.11

    Two fixes: default-initialization pollution from NaN-skipped elements in library and prediction sets, and missing string/utility includes that broke the macOS CRAN check. Build and edge-case hygiene between feature releases.

    View source ↗
  3. 6mo ago

    Spatially convergent partial cross mapping reaches the R API

    Another EDM method graduates from internal C++ to a documented R-level API with a vignette — the same pattern as the spatial logistic map in 1.7 and geographical pattern causality in 1.9. The remainder is console output trimming, include-order enforcement and optional white noise in the logistic map simulation.

    View source ↗
  4. 8mo ago

    Raster cross mapping with anisotropic embedding

    The heaviest recent release: geographical pattern causality gets its R API and vignette, and cross mapping extends to raster data with anisotropic embedding and variable sliding-window steps, plus mean-based and stacked spatial cross-sectional embeddings. Two breaking changes accompany it — internal case data removed and parameter order harmonised across the vector and raster S4 generics.

    View source ↗
  5. 11mo ago

    Configurable distance metrics and multithreaded distance computation

    Cross mapping gains configurable distance metrics and alternative spatial cross-sectional embedding styles, with multithreading added to distance-related computation for runtime. The same configurable-distance-metric change appears in sibling package tEDM within a week, confirming the shared core.

    View source ↗
  6. 1y ago

    Spatial logistic map exposed at the R level

    The spatial logistic map gets an R-level API and vignette, establishing the release pattern the following versions repeat. Four breaking changes ride along, including trend.rm giving way to detrend and a parameter rename in multiview().

    View source ↗