The list below tracks features actively planned for the fastsem engine and the R interface. Priorities are influenced by issues filed against fastsemR and fastsem.
Engine
- macOS binaries (arm64 and x86-64) using the Apple Accelerate framework for BLAS / LAPACK.
- Python bindings via nimpy, with a thin layer for JAX / PyTorch interoperability.
- Full GPU FIML gradient for large-N models with definition variables, lifting the current CPU-side gradient bottleneck.
- Additional fit indices — RMSEA, CFI, TLI, SRMR — alongside the existing chi-square and information criteria.
- Bayesian SEM with an HMC / NUTS posterior sampler.
- VCF streaming for biobank-scale Gen-SEM scans that exceed current memory limits.
R package (fastsemR)
- Bundled benchmark suite so users can reproduce the comparison numbers on their own hardware.
-
Tighter
umx::umxSummary()integration for derived parameters and standardised effects. - Convenience helpers for common Gen-SEM workflows (PLINK input, per-SNP result tables).
- CRAN release once the engine binary ships for all three major platforms.
Have a request?
Open an issue at https://github.com/lf-araujo/fastsemR/issues or, for engine-side questions, https://github.com/lf-araujo/fastsem/issues.