fastsemR

📚 Escolha o idioma · 选择语言 · Choose language
🇧🇷 Português · 🇨🇳 中文 · 🇬🇧 English

fastsemR is an R interface to fastsem, a compiled Nim SEM engine that supports:

• ML (full-data maximum likelihood) with analytical gradient and optional GPU acceleration
• FIML (full-information ML for missing data and definition variables)
• DWLS / WLSMV (diagonally weighted least squares for ordinal indicators)
• Multi-group models with configural or constrained parameters
• Equality constraints, parameter bounds, and derived parameters (delta-method SEs)
• Cluster-robust standard errors
• Standardised estimates (StdAll) with SEs via the delta method

The package translates umx / OpenMx RAM model objects into fastsem lavaan syntax, fits the model, and injects estimates back into the model object so that summary(), omxGetParameters(), and umxCompare() continue to work.

---

Installation

# Install from GitHub (requires devtools)
devtools::install_github("lf-araujo/fastsemR")

The compiled shared library is downloaded automatically from GitHub Releases the first time the package is loaded. To force a re-download (e.g. after a fastsem update):

library(fastsemR)
fastsem_update()

If you have built the library yourself from the fastsem source tree (nimble buildRnim), point the package to your local build:

fastsem_load("~/fastsem/libfastsem_r.so")

---

Example: mediation

Decompose the effect of engine displacement (disp) on fuel economy (mpg) through vehicle weight (wt). Specify the model in umx (or OpenMx) as usual, then let fastsem fit it: run_fastsem() returns the same MxModel, so every umx/OpenMx tool keeps working — including algebras such as the indirect effect ab, with delta-method SEs.

library(fastsemR)
library(umx)

df <- as.data.frame(scale(mtcars[, c("mpg", "disp", "wt")]))

m <- umxRAM("Mediation",
  umxPath("disp", to = "wt",  labels = "a"),
  umxPath("wt",   to = "mpg", labels = "b"),
  umxPath("disp", to = "mpg", labels = "c_prime"),
  umxPath(var   = c("disp", "wt", "mpg")),
  umxPath(means = c("disp", "wt", "mpg")),
  mxAlgebra(a * b, name = "ab"),
  data = df, autoRun = FALSE
)

m_fit <- run_fastsem(m)
umxSummary(m_fit)        # Algebra 'ab' = -0.483 CI95[-0.808, -0.158]
mxSE(ab, m_fit)          # 0.166, identical to mxRun()

The same works from lavaan: build the model with do.fit = FALSE and run_fastsem() returns a fitted lavaan object.

library(lavaan)
fit <- run_fastsem(sem("wt ~ a*disp\nmpg ~ b*wt + c*disp\nab := a*b",
                       data = df, do.fit = FALSE))
summary(fit)

run_fastsem() checks its own work: OpenMx / lavaan re-evaluate the model at the fastsem estimates, and a −2lnL that does not match the engine's is reported.

Contributing

Contributions are welcome. Please open an issue at https://github.com/lf-araujo/fastsemR/issues for bug reports, feature requests, or questions about the R interface. For issues specific to the underlying Nim engine (numerical results, GPU kernels, performance), file at https://github.com/lf-araujo/fastsem/issues.

Pull requests targeting main should pass R CMD check with no errors or warnings; the GitHub Actions workflow runs the same checks on Linux, macOS, and Windows.

---

License

fastsemR is released under the MIT License. See LICENSE for details.