R surface¶

This site's API reference (auto-generated from docstrings) covers only the Python frontend. The R frontend (package sezgi, in r-sezgi/) mirrors the Python class-first surface 1:1 by name, and is documented separately through R's own tooling:

109 roxygen-generated .Rd pages under r-sezgi/man/, covering every exported symbol.
r-sezgi/DESCRIPTION for the package's own summary and system requirements (SystemRequirements: Cargo (Rust's package manager), rustc (>= 1.88)).
Install with install.packages("sezgi", repos = c("https://tdelphi1981.r-universe.dev", "https://cloud.r-project.org")) (see Install → R for the full path, including the CRAN-submission note and the source-build fallback), then view any page from an R session:

library(sezgi)
?GeneticAlgorithm
?Algorithm

A browsable HTML reference site for the R package (e.g. via pkgdown) is not built by this milestone — it is a deliberate scope cut to avoid the R-side scope creep of authoring r-sezgi/README.md and vignettes from scratch (both of which pkgdown needs and neither of which exists today). This page is the honest summary in the meantime: it does not claim a generated R site exists.

Correspondence table¶

R6 classes and functions mirror their Python counterparts by name (sz_ prefix on free functions, matching the C-style prefix R packages conventionally use; class names are identical). Cross-language runs sharing a (spec/preset, problem, seed, budget) are bit-exact — verified by the project's own R/Python parity test suites, not merely asserted.

Python	R	Notes
`sezgi.Problem` (ABC)	`sezgi::Problem` (`R6::R6Class`)	Same `space()`/`evaluate()`/`optimum()` contract.
`sezgi.Algorithm`	`sezgi::Algorithm`	Engine-hosted `generate(pop, ctx)` base.
`sezgi.PopulationAlgorithm`	`sezgi::PopulationAlgorithm`	`select()`/`vary()` split.
`sezgi.LocalSearch`	`sezgi::LocalSearch`	`neighbor()`/`accept()` split.
`sezgi.AskTellAlgorithm`	`sz_algorithm()` / `sz_algo_solve()`	`setup()`/`step()` ask/tell surface, expressed as base-R functions (scope ruling: no R6 here).
`sezgi.GeneticAlgorithm`, ..., 29 built-in classes	`sezgi::GeneticAlgorithm`, ..., 29 built-in classes	One R6 class per Python wrapper class, table-driven on both sides from the same `crates/components/src/presets.rs` builder set.
`sezgi.solve(spec, problem, ...)`	`sz_solve_bbob()`, `sz_solve_onemax()`, `sz_solve_tsp()`, `sz_solve_cec2014()`/`cec2017()`/`cec2022()`, `sz_solve_cat_match()`, `sz_solve_int_quadratic()`, `sz_solve_mixed_diagnostic()`	Per-problem compat internals (see Solve / compat internals).
`sezgi.presets.*`	`sz_preset_*`	Same preset catalog, one builder per algorithm variant.
`sezgi.run_experiment`	`sz_run_experiment`	Same `ExperimentSpec`/IOH-logging semantics.
`sezgi.stats.*`	`sz_stats_*` (via `R/stats.R`)	Same statistical-comparison suite.
`sezgi.bias.*`	`sz_bias_*`	Same structural/central bias scanner.
`sezgi.mo.*`	`sz_nsga2`, `sz_mo_hypervolume_2d`, `sz_mo_igd`, `sz_mo_pareto_front`, ...	Same NSGA-II runner and MO indicators, bit-exact against the Python bindings.
`sezgi.Float`/`Int`/`Categorical`/`Binary`/`Permutation`/`Space`	`sz_float`/`sz_int`/`sz_categorical`/`sz_binary`/`sz_permutation`/`sz_space`	Same space builders.
`sezgi.recipes.FeatureSelection`/`MixedTuning`	`sezgi::FeatureSelection`/`MixedTuning`	Same two data recipes.

No overclaim¶

This page states what exists (109 .Rd pages, a class-for-class R6 mirror, bit-exact cross-language runs) and what does not (a generated, browsable pkgdown site; R vignettes). Neither is invented here.