Package {sezgi}

Title:	Metaheuristic Optimization with a 'Rust' Core
Version:	0.1.1
Description:	Build and run metaheuristic optimization algorithms as serializable component graphs executed by a 'Rust' core, with reproducible, bit-exact trajectories shared across the 'R' and 'Python' frontends. Includes population and local-search algorithm presets, standard benchmark suites (BBOB, CEC 2014/2017/2022, TSP), multi-objective indicators, structural-bias diagnostics and statistical comparison tools.
License:	MIT + file LICENSE
Copyright:	sezgi authors, except as documented in inst/COPYRIGHTS (bundled benchmark data and quoted reference excerpts)
Encoding:	UTF-8
Depends:	R (≥ 4.2)
Imports:	R6 (≥ 2.4.0)
Suggests:	testthat (≥ 3.0.0), jsonlite
SystemRequirements:	Cargo (Rust's package manager), rustc (>= 1.88)
URL:	https://github.com/tdelphi1981/sezgi
BugReports:	https://github.com/tdelphi1981/sezgi/issues
Config/testthat/edition:	3
Config/roxygen2/version:	8.1.0
Config/sezgi/MSRV:	1.88.0
NeedsCompilation:	yes
Packaged:	2026-09-05 13:03:59 UTC; tolgaberber
Author:	Tolga Berber [aut, cre], Beyzanur Siyah [aut], Emir Karayağız [aut], The authors of the dependency Rust crates [ctb] (see inst/AUTHORS file for details)
Maintainer:	Tolga Berber <tolga.berber@fen.ktu.edu.tr>
Repository:	CRAN
Date/Publication:	2026-09-15 10:40:02 UTC

Subclassable R6 base for an engine-hosted algorithm.

Description

Mirrors 'sezgi.Algorithm' ('py-sezgi/python/sezgi/algorithm.py', M4-1 Task 3) closely: a subclass implements 'generate(pop, ctx)' (REQUIRED) and may optionally override 'initialize_population(n, space, ctx)' (RULING 8: named 'initialize_population', NOT 'initialize' – R6 reserves ‘initialize' for the constructor, so Python’s 'initialize(n, ctx)' hook name cannot be reused here; a deliberate, documented divergence) and 'validate_space(space)'. 'run()' is the sole entry point, wiring the instance through T3's 'sz_solve_r_generator'/ 'sz_solve_r_generator_bbob' bridge.

Details

‘pop' (handed to 'generate()'/a 'PopulationAlgorithm'’s 'select()'/ ‘vary()'/a 'LocalSearch'’s 'neighbor()') is 'list(x, f)': 'f' is a numeric vector, the current population's fitness values, one entry per individual. 'x' is an n x dim numeric MATRIX (rows = individuals) for a SINGLE-Float-block space (the ergonomic, common case – 'pop$x[i, ]', 'ncol(pop$x)'); for every OTHER space shape (multi-block, or a single NON-Float block – Int/Categorical/Binary/Permutation), 'x' is instead a plain (unnamed) 'list' of per-individual values, one entry per individual (bare for a single block, a further per-block 'list' for multiple blocks) – the SAME single-vs-multi-block convention ‘Problem$evaluate(x)'’s own 'x' argument already uses. A 'generate()'/'vary()'/'neighbor()' return value mirrors this: a matrix (only meaningful when 'pop$x' was itself a matrix) or a plain 'list' of offspring x-values.

'ctx' is an 'environment' with '$rng' (an already-wrapped RNG handle – 'ctx$rng$next_f64()' (a double in '[0, 1)'), 'ctx$rng$next_below(n)' (an integer-valued double in '[0, n)'), 'ctx$rng$split(child_id)' (an independent child stream)), ‘$iteration' (a double, the engine’s own 0-based generation counter – always '0' for an 'initialize_population()' call), and '$space' (the block-descriptor 'list' for the space being solved – one entry per block, each a named 'list' with a 'type' field plus that block's own fields; the SAME shape ‘validate_space(space)'’s own 'space' argument uses).

Override-detection caveat: 'initialize_population'/'validate_space' are detected as overridden by comparing the resolved method against this class's own declared default at the CLASS level (body + formals, ignoring environment – see ‘R/algorithm.R'’s own module doc for the exact spelling) – a subclass method whose body happens to be byte-identical to the base default (e.g. a copy-pasted 'stop(...)' call with the same message) is therefore treated as NOT overridden even though it was technically redeclared; harmless today since both defaults are no-op-equivalent (‘initialize_population'’s default always errors if ever reached, ‘validate_space'’s default always accepts), but worth knowing.

See this file's own module doc (top of 'R/algorithm.R') for the full rationale behind these choices and the 'log_dir' rejection.

Methods:

'generate(pop, ctx)'

REQUIRED: produce this generation's offspring. Default: 'stop()'s with "not implemented".

'initialize_population(n, space, ctx)'

Optional: produce the initial population of size 'n'. Default: NOT overridden – 'run()' never calls this default body; the engine's own Rust 'init/uniform' initializer runs instead, with NO R call for population initialization at all (byte-identical to a raw 'sz_solve_r_generator' call with 'initializer = NULL').

'validate_space(space)'

Optional build-time veto: 'stop()' to reject 'space' BEFORE any 'generate()'/'initialize_population()' call. Default: no-op (accepts any space).

'run(problem, budget, pop_size = 50, seed = 0, run_id = 0, name = NULL, log_dir = NULL)'

Runs this algorithm's hooks INSIDE the Rust engine loop. 'problem': a 'Problem' subclass instance (routed through 'sz_as_problem()') OR a built-in problem descriptor (see [sz_builtin_bbob()]). 'log_dir': NOT supported (see module doc) – a non-'NULL' value raises a clear error. Returns an 'sz_result' (see ['.sz_wrap_result']). NOTE – known, INTENTIONAL divergence from 'sezgi.Algorithm.run(problem, budget, seed=0, pop_size=50, log_dir=None, ...)' (py-sezgi/python/sezgi/algorithm.py): 'pop_size' and 'seed' are swapped, and 'log_dir' sits last rather than third, kept as-is (not reordered to match) because every anchored test and example already calls 'run()' positionally against THIS order – reshuffling would break them for no benefit in a language where both sides are typically called with named arguments anyway.

Methods

Public methods

• Algorithm$generate()

• Algorithm$initialize_population()

• Algorithm$validate_space()

• Algorithm$run()

• Algorithm$clone()

Algorithm$generate()

Usage

Algorithm$generate(pop, ctx)

Algorithm$initialize_population()

Usage

Algorithm$initialize_population(n, space, ctx)

Algorithm$validate_space()

Usage

Algorithm$validate_space(space)

Algorithm$run()

Usage

Algorithm$run(
  problem,
  budget,
  pop_size = 50,
  seed = 0,
  run_id = 0,
  name = NULL,
  log_dir = NULL
)

Algorithm$clone()

The objects of this class are cloneable with this method.

Usage

Algorithm$clone(deep = FALSE)

Arguments

Examples

RandomStep <- R6::R6Class("RandomStep", inherit = Algorithm, public = list(
  generate = function(pop, ctx) {
    step <- vapply(seq_len(ncol(pop$x)), function(i) ctx$rng$next_f64() - 0.5, numeric(1))
    matrix(pop$x[1, ] + step, nrow = 1)
  }
))
res <- RandomStep$new()$run(sz_builtin_bbob(1, 2, 1), budget = 100, pop_size = 1, seed = 1)
res$evals_used

Differential Evolution – delegates to 'sz_preset_de_rand_1' (default, DE/rand/1/bin), 'sz_preset_de_best_1' (DE/best/1/bin), or 'sz_preset_jde' (self-adaptive jDE), selected by 'variant' (one of '"rand_1"', '"best_1"', '"jde"') at '$new()' time. All three presets share the identical '(pop_size, budget)' signature, so 'variant' is the only dispatch axis – no problem introspection needed (unlike [GeneticAlgorithm]'s space-driven auto-dispatch).

Description

See [sz_preset_classes]'s own doc for the problem-form support matrix and result shape (identical here).

Methods

Public methods

• DifferentialEvolution$new()

• DifferentialEvolution$run()

• DifferentialEvolution$clone()

DifferentialEvolution$new()

Usage

DifferentialEvolution$new(pop_size = 20, variant = "rand_1", ...)

DifferentialEvolution$run()

Usage

DifferentialEvolution$run(problem, budget, seed = 0, run_id = 0)

DifferentialEvolution$clone()

The objects of this class are cloneable with this method.

Usage

DifferentialEvolution$clone(deep = FALSE)

Arguments

The ask/tell evaluation session – see the module doc.

Description

The ask/tell evaluation session – see the module doc.

Usage

EvalSession

Binary feature-selection search over a 2D dataset's columns.

Description

Mirrors 'sezgi.recipes.FeatureSelection' ('py-sezgi/python/sezgi/recipes.py') exactly.

Details

'space()' is 'sz_space(sz_binary(n_features))': a genotype is a length-‘n_features' logical mask (per [Problem]’s own genotype conversion table – a Binary block converts to a logical vector), 'TRUE' selecting a column.

MINIMIZE convention: 'scorer(X_sub, y)' must return a LOWER-is-better number (e.g. an error/loss/residual metric, NOT accuracy/R^2/any higher-is-better score) – 'evaluate()' never negates or inverts it. An sklearn-shaped cross-validated ACCURACY scorer would need to be wrapped to flip its sign before it could be used here (this package has no ML-package dependency of its own – the sketch below is a COMMENT only, never an actual call anywhere in this file):

# scorer <- function(X_sub, y) {
#   # a cross-validated accuracy is HIGHER-is-better -- negate it so
#   # lower is better, matching this class's minimize convention.
#   -mean(cross_val_accuracy(X_sub, y))
# }
# FeatureSelection$new(X, y, scorer, penalty = 0.01)

'evaluate(mask)' = 'scorer(X[, mask, drop = FALSE], y) + penalty * (popcount(mask) / n_features)' – the penalty term is a fraction of the FULL feature count ('popcount / n_features', not a raw popcount), so it stays on a comparable scale to ‘scorer'’s own output regardless of 'n_features', and its size is 'penalty' at the all-features mask, '0' at the empty mask. Larger 'penalty' biases the search toward smaller feature subsets; 'penalty = 0' (default) is a pure 'scorer'-value search with no feature-count preference of its own.

'X[, mask, drop = FALSE]': the 'drop = FALSE' is LOAD-BEARING – without it, a single-‘TRUE' mask would subset 'X' down to a plain vector (R’s single-bracket auto-drop for a one-column result), silently changing ‘X_sub'’s shape from a matrix to a vector depending on the mask's own popcount and breaking any 'scorer' that assumes a consistent '(n_samples, k)' matrix shape (e.g. anything calling 'ncol(X_sub)' or doing matrix algebra on it) – the R analogue of M4-1 Task 6's Python bool-indexing trap, pinned here with its own regression test ('tests/testthat/test-oop-recipes.R').

EMPTY MASK ('popcount == 0'): 'scorer' is NEVER called – 'X[, mask, drop = FALSE]' would be an '(n_samples, 0)' matrix, and most real scorers (a model fit, a distance-correlation-like statistic, ...) cannot meaningfully score zero columns; the 'popcount == 0' case is instead handled directly, as a DOCUMENTED SENTINEL: 'evaluate(rep(FALSE, n_features)) == Inf'. 'Inf' was chosen over, say, a large-but-finite number because it is unambiguous (no dataset-dependent magnitude to pick or accidentally beat by a legitimately bad but non-empty selection) and it sorts correctly under every consumer's own comparison ('<') with no special-casing needed – the empty mask is simply never the minimizer of any run that has at least one non-empty candidate in its population, which every population-based search here always does.

Methods

'new(X, y, scorer, penalty = 0)'

'X' a 2D matrix/data.frame (coerced via 'as.matrix()'), 'y' the target (any vector/type 'scorer' accepts), 'scorer(X_sub, y) -> numeric(1)' a caller-supplied minimize-convention scoring function, 'penalty' a numeric scalar (default '0', no feature-count preference).

'space()'

'sz_space(sz_binary(n_features))', where 'n_features = ncol(X)'.

'evaluate(mask)'

See the class doc above.

Super class

Problem -> FeatureSelection

Methods

Public methods

• FeatureSelection$new()

• FeatureSelection$space()

• FeatureSelection$evaluate()

• FeatureSelection$clone()

FeatureSelection$new()

Usage

FeatureSelection$new(X, y, scorer, penalty = 0)

FeatureSelection$space()

Usage

FeatureSelection$space()

FeatureSelection$evaluate()

Usage

FeatureSelection$evaluate(mask)

FeatureSelection$clone()

The objects of this class are cloneable with this method.

Usage

FeatureSelection$clone(deep = FALSE)

Arguments

Genetic Algorithm – auto-dispatches to 'sz_preset_ga_real'/'ga_perm'/ ‘ga_bin'/'ga_int'/'ga_cat' ('R/000-wrappers.R') based on 'problem'’s own search space, read via 'prob$space()'/'.sz_space_to_blocks()':

Description

all-Float

'sz_preset_ga_real'

all-Permutation

'sz_preset_ga_perm'

all-Binary

'sz_preset_ga_bin'

all-Int

'sz_preset_ga_int'

all-Categorical

'sz_preset_ga_cat'

Details

A space MIXING block kinds (e.g. Float + Int together) has no 'ga_*' preset in this milestone – '$run()' raises a clear error naming the limitation (mirrors py-sezgi's ‘GeneticAlgorithm'’s own identical 'NotImplementedError', M4-1 Task 5 deferral (d)).

'representation' (a constructor kwarg, one of '"real"'/'"perm"'/ '"bin"'/'"int"'/'"cat"') OVERRIDES auto-dispatch entirely – block-kind introspection is skipped whenever it is given.

After a '$run()' call, '$dispatched_representation' holds the representation actually engaged – an introspectable field proving which preset ran, independent of 'representation' itself (which stays 'NULL' under auto-dispatch).

See [sz_preset_classes]'s own doc for the problem-form support matrix and result shape (identical here).

Methods

Public methods

• GeneticAlgorithm$new()

• GeneticAlgorithm$run()

• GeneticAlgorithm$clone()

GeneticAlgorithm$new()

Usage

GeneticAlgorithm$new(pop_size = 20, representation = NULL, ...)

GeneticAlgorithm$run()

Usage

GeneticAlgorithm$run(problem, budget, seed = 0, run_id = 0)

GeneticAlgorithm$clone()

The objects of this class are cloneable with this method.

Usage

GeneticAlgorithm$clone(deep = FALSE)

Arguments

Family base for single-trajectory local search (hill-climbing/ SA-shaped).

Description

Inherits [Algorithm]. A subclass implements 'neighbor(x, ctx)' (REQUIRED – proposes one perturbed candidate from the current point) and may optionally override 'accept(f_old, f_new, ctx)' (default: greedy, 'f_new <= f_old'). Designed for 'pop_size = 1' – 'run()' is overridden here to default AND ENFORCE 'pop_size = 1' ('stop()' otherwise).

Details

Mirrors 'sezgi.LocalSearch' ('py-sezgi/python/sezgi/algorithm.py', M4-1 Task 4) exactly, including its honest structural limitation:

'pop' (handed to 'generate()', which itself calls 'neighbor()'/ 'accept()') is 'list(x, f)', at 'pop_size = 1' always a single individual: ‘f' is a length-1 numeric vector, the current point’s fitness; 'x' is a length-1 x dim numeric MATRIX (a single row) for a SINGLE-Float-block space ('x[1, ]' is the current point), or a length-1 'list' otherwise ('x[[1]]' is the current point, bare for a single non-Float block, a further per-block 'list' for multiple blocks) – the SAME single-vs-multi-block convention 'Problem$ evaluate(x)''s own 'x' uses ('.sz_pop_at(pop, 1)', used internally by ‘generate()' below, abstracts over both shapes). 'neighbor()'’s return value is a SINGLE x-value in that same convention (NOT a list – 'generate()' wraps it). 'ctx' is an 'environment' with '$rng' (a wrapped RNG handle – '$rng$next_f64()'/'$rng$next_below(n)'/'$rng$ split(child_id)'), '$iteration' (a double), and '$space' (the block- descriptor ‘list' for the space being solved). See [Algorithm]’s own doc for the full detail on both, and for the override-detection caveat that applies to any 'initialize_population'/'validate_space' override this subclass adds (a body byte-identical to ‘Algorithm'’s own default is treated as NOT overridden, even if redeclared).

ACCEPT() DESIGN (read carefully – this is not the naive reading of "accept decides whether the engine keeps the point"): this base runs over the engine's default 'replace/mu-plus-lambda' replacer. At 'pop_size = 1' (mu = 1), with 'generate()' here always returning exactly ONE offspring (lambda = 1), that replacer ALWAYS keeps the strictly-better (or, on an exact tie, the OLD) of {current, neighbor} as the NEXT call's ‘pop$x'/'pop$f'’s own first entry – a structural, UNCONDITIONAL guarantee of 'replace/mu-plus-lambda' itself, not a decision 'accept()' makes. Two direct consequences: (1) the fitness this base reads on the call following a 'neighbor()' proposal is ALWAYS '<= f_old' – 'accept()' can therefore NEVER actually be called with an 'f_new' that is objectively WORSE than 'f_old'; a worse neighbor's exact fitness is never even surfaced back to R by this bridge. **This base's 'accept()' therefore CANNOT implement true simulated-annealing-style "sometimes accept a worse move" semantics – that would require intercepting the replacer's own decision, which this design deliberately does NOT attempt.** (2) Given (1), the DEFAULT 'accept' ('f_new <= f_old') is a TAUTOLOGY over every pair this base can ever actually present to it – it is ALWAYS true, faithfully mirroring (not fighting) what 'replace/mu-plus-lambda' already unconditionally enforces at 'pop_size = 1'. What 'accept()' DOES meaningfully control: whether this base's OWN internally-tracked anchor ('self$current_x'/'self$current_f', the point the NEXT 'neighbor()' call is proposed from) ADVANCES to match the engine's already-decided outcome, or stays where it was – i.e. ‘accept()' governs this base’s own SEARCH TRAJECTORY, never population membership (which the replacer alone decides, unconditionally, at 'pop_size = 1').

Super class

Algorithm -> LocalSearch

Methods

Public methods

• LocalSearch$neighbor()

• LocalSearch$accept()

• LocalSearch$run()

• LocalSearch$generate()

• LocalSearch$clone()

LocalSearch$neighbor()

Usage

LocalSearch$neighbor(x, ctx)

LocalSearch$accept()

Usage

LocalSearch$accept(f_old, f_new, ctx)

LocalSearch$run()

Usage

LocalSearch$run(
  problem,
  budget,
  pop_size = 1,
  seed = 0,
  run_id = 0,
  name = NULL,
  log_dir = NULL
)

LocalSearch$generate()

Usage

LocalSearch$generate(pop, ctx)

LocalSearch$clone()

The objects of this class are cloneable with this method.

Usage

LocalSearch$clone(deep = FALSE)

Arguments

The general "tune anything" door: binds a caller-supplied 'objective(x) -> numeric(1)' over ANY declared 'sz_space(...)' – a single block ('sz_float(...)', 'sz_categorical(...)', ...) or a multi-block space. 'evaluate(x)' delegates to 'objective' UNCHANGED – ‘x'’s exact shape follows [Problem]'s own genotype conversion table (a bare converted value for a single-block space, an unnamed 'list' of per-block converted values in ‘space()'’s own block order for a multi-block one).

Description

Mirrors 'sezgi.recipes.MixedTuning' ('py-sezgi/python/sezgi/recipes.py') exactly.

Details

Note [GeneticAlgorithm] auto-dispatches only over a SINGLE-kind space (all-Float, all-Binary, ...) – it raises a clear error naming the limitation on a genuinely Mixed space (see its own doc); a mixed-space 'MixedTuning' instance is instead run via a hand-built 'gen/compound' algorithm spec passed to 'sezgi:::sz_solve_r_problem()' directly (see ‘tests/testthat/test-oop-recipes.R'’s own worked example), or 'evaluate()'d directly for a non-engine use (grid search, a script sanity check, ...). A single-kind 'MixedTuning' space (e.g. Float-only, tuning several continuous hyperparameters at once) runs through [GeneticAlgorithm] exactly like any other single-kind [Problem].

Methods

'new(space, objective)'

'space' an 'sz_space(...)' object, 'objective(x) -> numeric(1)' a caller-supplied minimize-convention function.

'space()'

Returns the constructor's own 'space', unchanged.

'evaluate(x)'

Returns 'objective(x)', unchanged.

Super class

Problem -> MixedTuning

Methods

Public methods

• MixedTuning$new()

• MixedTuning$space()

• MixedTuning$evaluate()

• MixedTuning$clone()

MixedTuning$new()

Usage

MixedTuning$new(space, objective)

MixedTuning$space()

Usage

MixedTuning$space()

MixedTuning$evaluate()

Usage

MixedTuning$evaluate(x)

MixedTuning$clone()

The objects of this class are cloneable with this method.

Usage

MixedTuning$clone(deep = FALSE)

Arguments

Class skin over [sz_nsga2()] ('R/mo.R') – NOT one of the 34 'sz_preset_*' builders (NSGA-II is not built on the scalar Engine/Registry/Generator/AlgorithmSpec machinery every 'sz_preset_*' targets, see this file's own module doc). ZERO new MO capability: '$run()' delegates to 'sz_nsga2()' VERBATIM, same positional/keyword arguments, same return value shape.

Description

Constructor kwargs mirror ‘sz_nsga2()'’s own "algorithm configuration" parameters – 'pop_size' plus every variation-operator knob ('eta_c', 'eta_m', 'p_c', 'p_m', 'p_c_bin', 'p_m_bin', 'p_c_cat', 'p_m_cat'), the ones that describe the algorithm instance itself, independent of which problem it is pointed at. ‘$run()'’s own arguments mirror ‘sz_nsga2()'’s remaining "this particular problem/run" parameters ('problem', 'dim', 'budget', 'm', 'k', 'l', 'seed', 'log_dir', 'label') – 'm'/'k'/'l' describe the PROBLEM being solved ('m' = objective count for dtlz/wfg; ‘k'/'l' = WFG’s own shape parameters), not the algorithm, so they belong with ‘$run()' rather than '$new()', exactly mirroring 'sz_nsga2()'’s own per-problem-family validation.

Details

'NSGA2$new(pop_size = P, ...)$run(problem, dim, budget, m = M, seed = S, ...)' is IDENTICAL to calling 'sz_nsga2(problem, dim, P, budget, m = M, seed = S, ...)‘ directly (see 'test-oop-builtins.R'’s own anchored equivalence tests, a ZDT anchor and a WFG anchor where 'k'/'l' matter).

‘$run()'’s return value is ‘sz_nsga2()'’s OWN list shape ('individuals', 'objectives', 'front0', 'evals_used', 'violations' when constrained) – NOT an ‘sz_result' (T4’s '.sz_wrap_result()' shape assumes a single-objective run with one best point, which does not fit NSGA-II's multi-objective Pareto-front result).

Methods

Public methods

• NSGA2$new()

• NSGA2$run()

• NSGA2$clone()

NSGA2$new()

Usage

NSGA2$new(
  pop_size,
  eta_c = 20,
  eta_m = 20,
  p_c = 0.9,
  p_m = NULL,
  p_c_bin = 0.9,
  p_m_bin = NULL,
  p_c_cat = 0.9,
  p_m_cat = NULL
)

NSGA2$run()

Usage

NSGA2$run(
  problem,
  dim,
  budget,
  m = NULL,
  seed = 0,
  k = NULL,
  l = NULL,
  log_dir = NULL,
  label = NULL
)

NSGA2$clone()

The objects of this class are cloneable with this method.

Usage

NSGA2$clone(deep = FALSE)

Arguments

Family base for population-style algorithms (GA/DE/ES-shaped).

Description

Inherits [Algorithm]. A subclass implements 'vary(parents, ctx)' (REQUIRED – turns a parent pool into offspring) and may optionally override 'select(pop, k, ctx)' (default: k-fold binary tournament, tournament size 2, minimization). 'generate(pop, ctx)' is NOT meant to be overridden (override 'select'/'vary' instead) – it is composed here from 'select' + 'vary'.

Details

Mirrors 'sezgi.PopulationAlgorithm' ('py-sezgi/python/sezgi/ algorithm.py', M4-1 Task 4) exactly, including the PROVEN-uniform two-fixed-draw index-shift tournament formula (ported verbatim below, with its own determinism comment).

'pop' (handed to 'select()'/'generate()') is 'list(x, f)': 'f' is a numeric vector of per-individual fitness values; 'x' is an n x dim numeric MATRIX (rows = individuals) for a SINGLE-Float-block space, or a plain (unnamed) 'list' of per-individual values otherwise (bare for a single non-Float block, a further per-block 'list' for multiple blocks) – the SAME single-vs-multi-block convention 'Problem$ evaluate(x)''s own ‘x' uses. 'vary()'’s return value mirrors this (a matrix or a 'list'). 'ctx' is an 'environment' with '$rng' (a wrapped RNG handle – '$rng$next_f64()'/'$rng$next_below(n)'/'$rng$split (child_id)'), '$iteration' (a double), and '$space' (the block- descriptor ‘list' for the space being solved). See [Algorithm]’s own doc for the full detail on both, and for the override-detection caveat that applies to any 'initialize_population'/'validate_space' override this subclass adds (a body byte-identical to ‘Algorithm'’s own default is treated as NOT overridden, even if redeclared).

Methods:

'select(pop, k, ctx)'

Default: k-fold binary tournament selection (tournament size 2; minimization – lower fitness wins). For each of the 'k' requested parent slots, INDEPENDENTLY draws exactly two values from 'ctx$rng' (with 'n = ' the current population size): 'i <- ctx$rng$next_below(n)'; 'j <- ctx$rng$next_below(n - 1); if (j >= i) j <- j + 1' (for 'n > 1'). This is the standard "index-shift" trick for drawing two DISTINCT indices from '[0, n)' using exactly two draws and no rejection/ retry loop – the draw count per slot is always EXACTLY 2 for 'n > 1', so the whole sequence is hand-traceable given a fixed RNG seed and a known 'n'. The individuals at 'i' and 'j' then "fight": 'pop$f[i+1] <= pop$f[j+1]' wins as 'i' – this is also the EXACT tie-break rule: on an exact tie, the first-drawn contestant ('i') always wins, never 'j'. 'n == 1' (a population of one) is a degenerate edge case with no second index to draw: only 'i <- ctx$rng$next_below(1)' (always '0') is drawn, and 'i' wins trivially, costing 1 draw instead of 2. Returns a 'list' of exactly 'k' parents (individual x-values, one per slot, in slot order) – ties/duplicates across slots are possible and expected.

'vary(parents, ctx)'

REQUIRED: turn a parent pool (a 'list' of ‘length(parents)' x-values) into this generation’s offspring (a matrix or a 'list', see the module doc). Offspring COUNT is entirely this method's own choice, independent of 'length(parents)'.

'generate(pop, ctx)'

Composes 'select()' then 'vary()' – NOT meant to be overridden. Arity contract (PINNED): always calls ‘self$select(pop, n, ctx)' where 'n' is the CURRENT population’s own size (the standard "mating pool" convention: a full-size parent pool). The resulting 'parents' list is passed to 'self$vary(parents, ctx)‘ UNCHANGED, and 'vary()'’s return value is returned UNCHANGED as this generation's offspring.

Super class

Algorithm -> PopulationAlgorithm

Methods

Public methods

• PopulationAlgorithm$select()

• PopulationAlgorithm$vary()

• PopulationAlgorithm$generate()

• PopulationAlgorithm$clone()

PopulationAlgorithm$select()

Usage

PopulationAlgorithm$select(pop, k, ctx)

PopulationAlgorithm$vary()

Usage

PopulationAlgorithm$vary(parents, ctx)

PopulationAlgorithm$generate()

Usage

PopulationAlgorithm$generate(pop, ctx)

PopulationAlgorithm$clone()

The objects of this class are cloneable with this method.

Usage

PopulationAlgorithm$clone(deep = FALSE)

Arguments

Subclassable R6 base for a search-space problem.

Description

Mirrors 'sezgi.Problem' ('py-sezgi/python/sezgi/problem.py') exactly: subclass and override 'evaluate(x)' and 'space()' (both required – the defaults below 'stop()' with a "not implemented" message); 'optimum()' is an optional override, default 'NULL'.

Details

Genotype -> R conversion table for ‘evaluate(x)'’s own 'x' argument (PINNED; mirrors M4-1's pinned block -> Python table exactly, see ‘py-sezgi/python/sezgi/problem.py'’s own module doc):

Float block

a numeric (double) vector

Int block

an integer vector

Categorical block

an integer vector of category INDICES '0..k', not labels

Binary block

a logical vector

Permutation block

an integer vector, 0-based

A single-block space's ‘x' is that one block’s own converted value, passed BARE. A multi-block space's 'x' is an (unnamed) R 'list' of per-block converted values, in ‘space()'’s own block order.

‘batch_evaluate(xs)'’s own 'xs' is a 'list' of 'x' values (one per individual, each in the SAME per-‘x' shape as 'evaluate(x)'’s own 'x' above), matching py-sezgi's 'sezgi.Problem.batch_evaluate' input shape exactly ('py-sezgi/python/sezgi/problem.py') – the Python side is the authority for this shape, not an independent R design.

Methods:

'evaluate(x)'

'x' -> a numeric scalar, the fitness/objective value at ‘x'. Subclasses MUST override this. See the class doc’s conversion table above for ‘x'’s exact shape. Default: 'stop()'s with "not implemented".

'space()'

Declares the search space. Subclasses MUST override this to return an 'sz_space(...)' object. Default: 'stop()'s with "not implemented".

'optimum()'

The problem's known optimum (a numeric scalar), or 'NULL' (the default) if it has none. Optional override.

'batch_evaluate(xs)'

'xs' -> a numeric vector, one entry per 'x' in 'xs', in the SAME order (default: loop 'self$evaluate', mirroring ‘sezgi.Problem.batch_evaluate'’s own default '[self.evaluate(x) for x in xs]' exactly). Override for a vectorized objective; called ONCE PER GENERATION with the WHOLE population as ‘xs' (see '.sz_make_evaluate_shim'’s own doc).

Methods

Public methods

• Problem$evaluate()

• Problem$space()

• Problem$optimum()

• Problem$batch_evaluate()

• Problem$clone()

Problem$evaluate()

Usage

Problem$evaluate(x)

Problem$space()

Usage

Problem$space()

Problem$optimum()

Usage

Problem$optimum()

Problem$batch_evaluate()

Usage

Problem$batch_evaluate(xs)

Problem$clone()

The objects of this class are cloneable with this method.

Usage

Problem$clone(deep = FALSE)

Arguments

Examples

Sphere <- R6::R6Class("Sphere", inherit = Problem, public = list(
  space = function() sz_space(sz_float(-5, 5, 2)),
  evaluate = function(x) sum(x^2),
  optimum = function() 0
))
prob <- Sphere$new()
prob$evaluate(c(1, 2))

An owned per-call RNG handle exposed to an R-authored 'generate'/ 'initialize' callback as the 'rng' argument – an external-pointer- backed '#[savvy]' object ('SzRng$new'/'SzRng$from_master'/method closures are all auto-generated by 'savvy-cli', the SAME mechanism ‘EvalSession' already uses, 'session.rs'’s own module doc). Wraps a CLONE of the stage's live 'RngStream' – mirrors 'PyRng' ('py-sezgi/ src/lib.rs‘) exactly: every method here delegates 1:1 to 'RngStream'’s own 'pub' API ('crates/core/src/rng.rs:16-59'), no new RNG logic.

Description

**Clone-out / write-back / null-out protocol** (research doc §C5): ['RGenerator::generate']/['RInitializer::initialize'] clone '*ctx.rng' into a fresh 'SzRng' BEFORE the call, 'into_external_pointer()' it, and hand the raw pointer to R as one argument. After the call returns, the mutated state is read back via ['take_external_pointer_value'] – which hands back OWNERSHIP of the (possibly-advanced) 'SzRng' AND nulls the R-side pointer ('R_ClearExternalPtr') in the SAME call. This is a DELIBERATE IMPROVEMENT over an earlier stale-handle fix, which was docstring-only ("valid only for the duration of one 'generate()' call"): here, a callback that STORES 'ctx$rng' in its own enclosure across iterations gets a clear 'savvy::Error::InvalidPointer' ("This external pointer is already consumed or deleted") on any later use, rather than silently reading stale state – a hazard turned into an honest error, not just a documented one.

Usage

SzRng

Prints an 'sz_block' object (any of the five block kinds).

Description

Prints an 'sz_block' object (any of the five block kinds).

Usage

## S3 method for class 'sz_block'
print(x, ...)

Arguments

Value

'x', invisibly.

Prints an 'sz_result' object.

Description

Prints an 'sz_result' object.

Usage

## S3 method for class 'sz_result'
print(x, ...)

Arguments

Value

'x', invisibly.

Prints an 'sz_space' object.

Description

Prints an 'sz_space' object.

Usage

## S3 method for class 'sz_space'
print(x, ...)

Arguments

Value

'x', invisibly.

Version of the underlying sezgi Rust core.

Description

Version of the underlying sezgi Rust core.

Usage

sezgi_version()

Value

A character scalar with the crate version string.

Drives an 'sz_algorithm' over an 'EvalSession' to exhaustion.

Description

Semantics MIRROR py-sezgi's 'Algorithm.solve()' exactly ('py-sezgi/python/sezgi/algo.py'):

• ‘set.seed(seed)' is called ONCE, up front – R’s global RNG IS the ‘ctx' RNG for the whole run (see '.sz_algo_context()'’s doc); 'algo$setup'/'algo$step' calling 'ctx$random_point()' (or drawing from R's RNG directly, e.g. 'runif()'/'sample()') are both driven by this one seeded stream, so a run is fully reproducible from 'seed' alone.

• The driver calls 'algo$setup(ctx)' once, then 'algo$step(ctx)' repeatedly while 'ctx$remaining() > 0'. A 'step()' call that leaves 'ctx$evals_used()' UNCHANGED is a driver error ('stop()' with a message containing '"consumed no budget"') – a step must evaluate at least one point or let 'ctx$evaluate()' raise 'sz_budget_exhausted'.

• 'ctx$evaluate(points)' signals the condition class '"sz_budget_exhausted"' when 'points' would exceed the remaining budget – checked BEFORE the session is touched, so a rejected batch spends nothing. ‘sz_algo_solve()'’s own 'tryCatch' catches ONLY this condition class (wrapping both 'algo$setup()' and the 'step()' loop), ending the run cleanly; any OTHER error 'setup()'/'step()' raises (including the "consumed no budget" guard above) propagates out of 'sz_algo_solve()' uncaught.

• A run that ends having evaluated NOTHING at all ('ctx$best()' is 'NULL') errors.

• 'session$finish()' is GUARANTEED to run EXACTLY ONCE, on every exit path – success, the "consumed no budget"/"no evaluations" guards, or any error 'setup()'/'step()' itself raises – via 'on.exit(..., add = TRUE)' registered before the run starts (base R's ‘finally' equivalent). This mirrors py-sezgi’s own 'try'/ 'finally' structure, which a Task 4 (M3-4) final-review finding added specifically so a run that dies mid-flight does not silently lose its IOH archive; applied here from the start rather than rediscovered later.

Usage

sz_algo_solve(algo, session, seed)

Arguments

Value

A named 'list' with fields 'algo' (character, 'algo$name'), 'seed', 'budget' ('session$budget()'), 'evals_used', 'best_x' (numeric vector), 'best_f' (numeric scalar), 'f_opt' (numeric scalar, or 'NULL' if the problem has no known optimum – e.g. an f0 session), 'gap' ('best_f - f_opt', or 'NULL' iff 'f_opt' is 'NULL') – mirrors py-sezgi's 'SolveResult' field names exactly.

Declares a pure-R metaheuristic algorithm to run over an 'EvalSession'.

Description

Base R only (no R6/S4): the result is a plain classed 'list' holding the two closures and a name – there is no method dispatch, subclassing, or mutable state of its own; all run state lives in the 'ctx' environment ‘sz_algo_solve()' builds fresh for each call. Mirrors py-sezgi’s 'sezgi.Algorithm' ABC (subclass, implement 'setup()'/'step()'), expressed as two plain closures instead of two abstract methods.

Usage

sz_algorithm(setup, step, name = "custom")

Arguments

Value

A classed 'list' ('"sz_algorithm"') with elements 'setup', 'step', 'name'.

Accepts a 'Problem' subclass instance.

Description

Mirrors py-sezgi's 'as_native_problem()' ('py-sezgi/python/sezgi/problem.py') in spirit – unlike py-sezgi, r-sezgi builds no separate "native problem handle" type to convert to (there is no '_sezgi.Problem' counterpart), so a 'Problem' subclass instance already IS what this task's Rust bridge needs; this function's job is purely to give a friendly, dedicated error for the "not a Problem" case, at the point of use rather than deep inside the solve machinery.

Usage

sz_as_problem(obj)

Arguments

Value

'obj', unchanged, if it inherits '"Problem"'.

Bayesian Plackett-Luce posterior via Gibbs sampling.

Description

Mirrors py-sezgi's 'stats.bayesian_plackett_luce()'. Caron & Doucet (2012) latent exponential-race Gibbs augmentation under independent 'Gamma(1, 1)' priors; see 'crates/stats/src/bayesian.rs' for the full PINNED sampler and determinism guarantees. The same '(rankings, samples, burn_in, seed)' always produces a bit-identical result – R calls the exact same seeded Rust core as Python and Rust.

Usage

sz_bayesian_plackett_luce(rankings, samples = 2000, burn_in = 500, seed = 1)

Arguments

Value

A named list with 'mean_worths', 'ci_low', 'ci_high', 'p_best', 'samples'.

Central-bias scan: run an algorithm spec on paired centered/shifted BBOB conditions and test whether its performance gap differs between them (Kudela's center-bias-exploitation method). See 'crates/bias/src/central.rs' for the full method provenance.

Description

Central-bias scan: run an algorithm spec on paired centered/shifted BBOB conditions and test whether its performance gap differs between them (Kudela's center-bias-exploitation method). See 'crates/bias/src/central.rs' for the full method provenance.

Usage

sz_bias_central(
  spec_json,
  dim,
  budget,
  fids = NULL,
  instances_shifted = NULL,
  runs_per = 20,
  seed = 0
)

Arguments

Value

A named list with 'gap_centered', 'gap_shifted' (numeric vectors, same length/order – index 'i' in both is one matched pair), 'wilcoxon' (named list: 'w_statistic', 'z', 'p_value', 'n_effective', ‘method'), 'effect' (Cliff’s delta, 'gap_shifted' vs 'gap_centered'), ‘verdict', 'detail' – mirrors py-sezgi’s 'sezgi.bias.central()' dict keys exactly.

One-call bias report: runs both the structural and central bias scans on one algorithm spec and assembles a single report with a ready-to-paste LaTeX summary table. See 'crates/bias/src/report.rs' for the full assembly/LaTeX-rendering details.

Description

T6 (the Rajwar-Deep signature-bias test) is DEFERRED in the underlying 'sezgi-bias' crate itself – the method could not be pinned from accessible sources (no reachable source fully specifies its statistical procedure). There is no ‘sz_bias_signature()'; the returned list’s 'signature' element is mapped through properly (not hardcoded), so it is ‘NULL' today only because the crate’s own result is always 'NULL' today – see below – and 'latex_summary' renders an honest "not run" row for it instead of a fabricated result.

Usage

sz_bias_report(
  spec_json,
  dim,
  budget,
  seed = 0,
  structural_runs = NULL,
  central_fids = NULL,
  central_instances = NULL,
  central_runs_per = NULL
)

Arguments

Value

A named list with 'structural' (same shape as ‘sz_bias_structural()'’s return), 'central' (same shape as ‘sz_bias_central()'’s return), 'signature' ('NULL' today, see above; when non-'NULL', a two-element list 'list(verdict = ..., detail = ...)‘, same shape as 'structural'’s/‘central'’s own 'verdict'/'detail'), 'latex_summary' (character scalar, never contains the literal '"NaN"'), 'plot_data' (named list: 'final_positions', 'gap_centered', 'gap_shifted' – the same raw vectors already inside 'structural'/ ‘central') – mirrors py-sezgi’s 'sezgi.bias.report()' dict keys exactly.

Structural-bias scan: run an algorithm spec repeatedly on the f0 random-function null problem and test its final positions for departure from uniformity (BIAS-toolbox method; Kononova et al. 2015 / Vermetten et al. 2022). See 'crates/bias/src/structural.rs' for the full method provenance.

Description

Structural-bias scan: run an algorithm spec repeatedly on the f0 random-function null problem and test its final positions for departure from uniformity (BIAS-toolbox method; Kononova et al. 2015 / Vermetten et al. 2022). See 'crates/bias/src/structural.rs' for the full method provenance.

Usage

sz_bias_structural(spec_json, dim, budget, runs = 30, seed = 0)

Arguments

Value

A named list with 'per_dim_ks' (list of 'list(d=, p_value=, n=)', length 'dim'), 'per_dim_ad' (list of 'list(a2=, p_value=, n=)', same length/order), 'holm_rejections_ks', 'holm_rejections_ad', 'verdict' ('"no_evidence"' or '"evidence"'), 'detail' ('NULL' iff 'verdict == "no_evidence"', else a character scalar), 'final_positions' (list of 'runs' numeric vectors, each length 'dim') – mirrors py-sezgi's 'sezgi.bias.structural()' dict keys exactly.

Statistics-only structural-bias scan over externally-collected final positions – the bias bridge for algorithms authored OUTSIDE this package's own spec/engine (M3-5 Task 6), e.g. a pure-R sz_algorithm driven by sz_algo_solve over sz_eval_session_f0, one run at a time. Runs the SAME KS/AD/ Holm battery as sz_bias_structural over caller-supplied 'final_positions' instead of driving an algorithm spec through the engine itself – see crates/bias/src/structural.rs's ‘scan_from_positions' for the full method provenance. Mirrors py-sezgi’s 'sezgi.bias.structural_positions()' 1:1.

Description

Statistics-only structural-bias scan over externally-collected final positions – the bias bridge for algorithms authored OUTSIDE this package's own spec/engine (M3-5 Task 6), e.g. a pure-R sz_algorithm driven by sz_algo_solve over sz_eval_session_f0, one run at a time. Runs the SAME KS/AD/ Holm battery as sz_bias_structural over caller-supplied 'final_positions' instead of driving an algorithm spec through the engine itself – see crates/bias/src/structural.rs's ‘scan_from_positions' for the full method provenance. Mirrors py-sezgi’s 'sezgi.bias.structural_positions()' 1:1.

Usage

sz_bias_structural_positions(final_positions)

Arguments

Value

Same named-list shape as sz_bias_structural's return: 'per_dim_ks', 'per_dim_ad', 'holm_rejections_ks', 'holm_rejections_ad', 'verdict' ('"no_evidence"' or '"evidence"'), 'detail' ('NULL' iff 'verdict == "no_evidence"'), 'final_positions'.

A binary block: 'n' bits.

Description

Mirrors 'sezgi.Binary' ('py-sezgi/python/sezgi/spaces.py') exactly – see [sz_float()]'s doc for the no-construction-time-validation note.

Usage

sz_binary(n)

Arguments

Value

An object of class 'c("sz_block_binary", "sz_block")' with element 'n'.

A built-in-problem descriptor for [Algorithm]'s 'run()'.

Description

r-sezgi has no general "native problem handle" type the way py-sezgi's 'as_native_problem()' does (every existing 'sz_solve_*'/'sz_eval_session*' binding takes 'fid'/'dim'/'instance' as separate scalar arguments, never a wrapped object) – this is a MINIMAL, purpose-built descriptor letting ‘Algorithm$run()'’s single ‘problem' argument select T3’s 'sz_solve_r_generator_bbob()' entry point (the "built-in form" of a ‘run()' call, mirroring py-sezgi’s 'sezgi.bbob(fid, dim, instance)' handles reaching 'Algorithm.run()' the same way) instead of the 'Problem'-subclass path. Deliberately narrow: it exists ONLY to route 'Algorithm$run()', not as a general native-problem abstraction (a richer type, if ever needed by a future task, is out of this task's scope – see the task-4 report's concerns for T5).

Usage

sz_builtin_bbob(fid, dim, instance = 1)

Arguments

Value

An object of class '"sz_builtin_bbob"' with elements 'fid', 'dim', 'instance'.

A categorical block: 'n' genes, each a category index in '0..k'.

Description

Mirrors 'sezgi.Categorical' ('py-sezgi/python/sezgi/spaces.py') exactly – see [sz_float()]'s doc for the no-construction-time-validation note.

Usage

sz_categorical(k, n)

Arguments

Value

An object of class 'c("sz_block_categorical", "sz_block")' with elements 'k', 'n'.

Direct, one-shot evaluation of a CEC 2014 (Liang, Qu & Suganthan 2013) function at 'x', bypassing 'sz_solve_bbob'-style budget/engine machinery entirely – binds ['Cec2014::new'] + ['Cec2014::evaluate_batch'] exactly. Mirrors ‘sz_cec2022_evaluate' (M3-6 Task 10 – see py-sezgi’s 'sezgi.problems.cec2014_evaluate' for the identical binding on the Python side). See [‘Cec2014::new']’s own doc for the exact 'fid'/'dim' domain.

Description

Direct, one-shot evaluation of a CEC 2014 (Liang, Qu & Suganthan 2013) function at 'x', bypassing 'sz_solve_bbob'-style budget/engine machinery entirely – binds ['Cec2014::new'] + ['Cec2014::evaluate_batch'] exactly. Mirrors ‘sz_cec2022_evaluate' (M3-6 Task 10 – see py-sezgi’s 'sezgi.problems.cec2014_evaluate' for the identical binding on the Python side). See [‘Cec2014::new']’s own doc for the exact 'fid'/'dim' domain.

Usage

sz_cec2014_evaluate(fid, dim, x)

Arguments

Value

A numeric scalar.

# Errors A savvy error for 'fid' outside '1..=30', 'dim' outside '{10,30}', or 'length(x) != dim'.

The report's pinned 'F_i*' bias for a CEC 2014 function – binds ['Cec2014::f_star'] ('F_i* = 100*fid'). Does not depend on 'dim', so an internal probe 'dim = 10' is used purely to validate 'fid'.

Description

The report's pinned 'F_i*' bias for a CEC 2014 function – binds ['Cec2014::f_star'] ('F_i* = 100*fid'). Does not depend on 'dim', so an internal probe 'dim = 10' is used purely to validate 'fid'.

Usage

sz_cec2014_f_star(fid)

Arguments

Value

A numeric scalar.

# Errors A savvy error if 'fid' is outside '1..=30'.

Direct, one-shot evaluation of a CEC 2017 (Awad, Ali, Liang, Qu & Suganthan 2016) function at 'x', bypassing 'sz_solve_bbob'-style budget/engine machinery entirely – binds ['Cec2017::new'] + ['Cec2017::evaluate_batch'] exactly. Mirrors 'sz_cec2014_evaluate' (M3-6 Task 10 – see py-sezgi's 'sezgi.problems.cec2017_evaluate' for the identical binding on the Python side). See [‘Cec2017::new']’s own doc for the exact 'fid'/'dim' domain.

Description

Direct, one-shot evaluation of a CEC 2017 (Awad, Ali, Liang, Qu & Suganthan 2016) function at 'x', bypassing 'sz_solve_bbob'-style budget/engine machinery entirely – binds ['Cec2017::new'] + ['Cec2017::evaluate_batch'] exactly. Mirrors 'sz_cec2014_evaluate' (M3-6 Task 10 – see py-sezgi's 'sezgi.problems.cec2017_evaluate' for the identical binding on the Python side). See [‘Cec2017::new']’s own doc for the exact 'fid'/'dim' domain.

Usage

sz_cec2017_evaluate(fid, dim, x)

Arguments

Value

A numeric scalar.

# Errors A savvy error for 'fid' outside '{1} union {3..=30}', 'dim' outside '{10,30}', or 'length(x) != dim'. 'fid = 2' raises a dedicated error – the Rust ['sezgi_problems::Cec2017Error::Withdrawn'] message is surfaced VERBATIM, distinct from an ordinary out-of-range 'fid'.

The report's pinned 'F_i*' bias for a CEC 2017 function – binds ['Cec2017::f_star'] ('F_i* = 100*fid', the fid-gapped C dispatch bias, not the report's contiguous renumbering – see ‘Cec2017::new'’s own module doc). Does not depend on 'dim', so an internal probe 'dim = 10' is used purely to validate 'fid'.

Description

The report's pinned 'F_i*' bias for a CEC 2017 function – binds ['Cec2017::f_star'] ('F_i* = 100*fid', the fid-gapped C dispatch bias, not the report's contiguous renumbering – see ‘Cec2017::new'’s own module doc). Does not depend on 'dim', so an internal probe 'dim = 10' is used purely to validate 'fid'.

Usage

sz_cec2017_f_star(fid)

Arguments

Value

A numeric scalar.

# Errors A savvy error if 'fid' is outside '{1} union {3..=30}' ('fid = 2' included, via the ['sezgi_problems::Cec2017Error::Withdrawn'] message).

Direct, one-shot evaluation of a CEC 2022 function at 'x', bypassing 'sz_solve_bbob'-style budget/engine machinery entirely – binds ['Cec2022::new'] + ['Cec2022::evaluate_batch'] exactly. See ‘Cec2022::new'’s own doc for the exact 'fid'/'dim' domain.

Description

Direct, one-shot evaluation of a CEC 2022 function at 'x', bypassing 'sz_solve_bbob'-style budget/engine machinery entirely – binds ['Cec2022::new'] + ['Cec2022::evaluate_batch'] exactly. See ‘Cec2022::new'’s own doc for the exact 'fid'/'dim' domain.

Usage

sz_cec2022_evaluate(fid, dim, x)

Arguments

Value

A numeric scalar.

# Errors A savvy error for 'fid' outside '1..=12', 'dim' outside '{2,10,20}', 'dim = 2' for a hybrid function, or 'length(x) != dim'.

The report's pinned 'F_i*' bias for a CEC 2022 function – binds [‘Cec2022::f_star'] (module doc section 1.2’s table). 'f_star' does not depend on 'dim', so an internal probe 'dim = 10' is used purely to validate 'fid' (every 'fid' in '1..=12' accepts 'dim = 10', hybrids included).

Description

The report's pinned 'F_i*' bias for a CEC 2022 function – binds [‘Cec2022::f_star'] (module doc section 1.2’s table). 'f_star' does not depend on 'dim', so an internal probe 'dim = 10' is used purely to validate 'fid' (every 'fid' in '1..=12' accepts 'dim = 10', hybrids included).

Usage

sz_cec2022_f_star(fid)

Arguments

Value

A numeric scalar.

# Errors A savvy error if 'fid' is outside '1..=12'.

Exports the IOH archive at 'log_root' as a COCO/BBOB "old format" archive rooted at 'out_dir' – see 'sezgi_bench::coco_export'. Returns the list of written file paths (as strings), sorted for determinism.

Description

BBOB-only: COCO's "old format" IS the BBOB archive format and has no CEC counterpart, so this errors (naming the offending suite) if 'log_root' holds any non-BBOB scenario, rather than silently merging it into a 'bbob'-labeled archive. A mixed BBOB+CEC tree must be filtered to its BBOB records before exporting.

Usage

sz_coco_export(log_root, out_dir)

Arguments

Value

A character vector of written file paths, sorted.

ECDF/anytime curve(s) over an on-disk IOH archive.

Description

Mirrors py-sezgi's 'ecdf()'. See 'crates/bench/src/anytime.rs' for the full PINNED definitions ('default_targets', the ECDF denominator convention, 'hit_time').

Usage

sz_ecdf(log_root, targets = NULL, per_algo = TRUE)

Arguments

Value

A named list (see 'per_algo').

Start an ask/tell evaluation session over a BBOB problem.

Description

The session is the sole keeper of evaluation, tamper-proof counting, best-tracking and (optionally) IOH logging, driven entirely by candidate points an EXTERNAL algorithm supplies – the spec's engine-inside-out promise. Mirrors py-sezgi's 'sezgi.EvalSession'.

Usage

sz_eval_session(
  fid,
  dim,
  instance,
  budget,
  log_dir = NULL,
  algo_name = "custom",
  seed = 0
)

Arguments

Details

The returned object is a reference (external-pointer) handle, not a value: assigning it to another variable (e.g. s2 <- s) does NOT copy the session – both names alias the SAME mutable session, so a $evaluate() through either one advances the same counter and best. This is standard R external-pointer/environment behavior (the object literally IS an environment), not something specific to this package.

The returned object exposes:

• $evaluate(x) – batch-evaluates 'x' (a numeric matrix with rows = points and 'dim' columns, or a 'list' of 'dim'-length numeric vectors) and returns a numeric vector of 'f' values, one per point. All-or-nothing: on any error (dimension mismatch, a non-finite coordinate, or a batch that would cross the remaining budget) nothing is counted.

• $evals_used() – number of evaluations counted so far.

• $budget() – the session's total evaluation budget.

• $best() – list(x = <numeric vector>, f = <numeric scalar>) for the best evaluation seen so far, or NULL if nothing has been evaluated yet.

• $f_opt() – the problem's known optimum value, or NULL if it has none (e.g. an f0 session – see sz_eval_session_f0).

• $dim() – the search space's dimensionality.

• $bounds() – a length-2 numeric vector c(lo, hi), the uniform bounds of this session's continuous space (errors for a non-uniform/non-float space; unreachable through any constructor this package exposes today).

• $finish() – flushes the IOH log (if enabled) and marks the session finished; every method call afterward, including a second $finish(), raises an error containing "session finished".

$dim(), $bounds(), and every method above except $f_opt() behave identically regardless of which constructor built the session (sz_eval_session(), sz_eval_session_cec2022, or sz_eval_session_f0).

Value

An 'EvalSession' object (see above).

Start an ask/tell evaluation session over a CEC 2014 problem.

Description

The generic-session counterpart to sz_eval_session (which is BBOB-only), mirroring sz_eval_session_cec2022 exactly: builds a session over sz_cec2014_evaluate's same Cec2014 problem, with IOH logging allowed. Session metadata (suite = "sezgi-cec2014", name = "cec2014-f<fid>", instance = 1, f_opt = the function's pinned F_i* = 100*fid bias) matches py-sezgi's EvalSession.for_problem exactly for a CEC 2014 problem handle, so an IOH record produced by an R session and one produced by an equivalent Python session reconstruct to the identical (suite, fid, name, instance) key and f_opt.

Usage

sz_eval_session_cec2014(
  fid,
  dim,
  budget,
  log_dir = NULL,
  algo_name = "custom",
  seed = 0
)

Arguments

Details

See sz_eval_session for the full list of methods the returned object exposes (identical here, including $dim() / $bounds()).

Value

An 'EvalSession' object (see sz_eval_session).

Start an ask/tell evaluation session over a CEC 2017 problem.

Description

The generic-session counterpart to sz_eval_session (which is BBOB-only), mirroring sz_eval_session_cec2014 exactly: builds a session over sz_cec2017_evaluate's same Cec2017 problem, with IOH logging allowed. Session metadata (suite = "sezgi-cec2017", name = "cec2017-f<fid>", instance = 1, f_opt = the function's pinned F_i* = 100*fid bias) matches py-sezgi's EvalSession.for_problem exactly for a CEC 2017 problem handle.

Usage

sz_eval_session_cec2017(
  fid,
  dim,
  budget,
  log_dir = NULL,
  algo_name = "custom",
  seed = 0
)

Arguments

Details

fid = 2 ("Sum of Different Powers") was officially withdrawn from the CEC 2017 suite: this constructor raises the dedicated sezgi_problems::Cec2017Error::Withdrawn message VERBATIM for it, distinct from an ordinary out-of-range fid.

See sz_eval_session for the full list of methods the returned object exposes (identical here, including $dim() / $bounds()).

Value

An 'EvalSession' object (see sz_eval_session).

Start an ask/tell evaluation session over a CEC 2022 problem.

Description

The generic-session counterpart to sz_eval_session (which is BBOB-only): builds a session over sz_cec2022_evaluate's same Cec2022 problem, with IOH logging allowed (unlike sz_eval_session_f0, whose problem has no known optimum for a log archive to record). Session metadata (suite = "sezgi-cec2022", name = "cec2022-f<fid>", instance = 1, f_opt = the function's known optimum) matches py-sezgi's EvalSession.for_problem exactly for a CEC 2022 problem handle, so an IOH record produced by an R session and one produced by an equivalent Python session reconstruct to the identical (suite, fid, name, instance) key and f_opt.

Usage

sz_eval_session_cec2022(
  fid,
  dim,
  budget,
  log_dir = NULL,
  algo_name = "custom",
  seed = 0
)

Arguments

Details

See sz_eval_session for the full list of methods the returned object exposes (identical here, including $dim() / $bounds()).

Value

An 'EvalSession' object (see sz_eval_session).

Start an ask/tell evaluation session over the f0 BIAS-toolbox null problem.

Description

f0 (sezgi_bias::F0Random on the Rust side) has no landscape at all: every evaluation is an independent U(0,1) draw over the domain ⁠[0,1]^dim⁠, uncorrelated with the point being queried. It exists purely as a probe for an algorithm's OWN structural bias – see sz_bias_structural.

Usage

sz_eval_session_f0(dim, f0_seed, budget)

Arguments

Details

Deliberately has NO log_dir/algo_name/seed (log) arguments: f0 has no known optimum, and IOH logging requires one (see sz_eval_session's log_dir for the BBOB/CEC-2022 case). $f_opt() on the returned session always returns NULL.

See sz_eval_session for the full list of methods the returned object exposes (identical here, including $dim() / $bounds(), which returns c(0, 1)).

Value

An 'EvalSession' object (see sz_eval_session).

Start an ask/tell evaluation session over a vendored TSPLIB (TSP) instance.

Description

M3-8 Task 8: the PERMUTATION-typed counterpart to sz_eval_session and its continuous siblings – the R mirror of py-sezgi's M3-8 Task 7 sezgi.EvalSession.for_problem(sezgi.problems.tsp(...)) typed session. $kind() on the returned session is "permutation" (every other constructor in this file returns a "float"-kind session).

Usage

sz_eval_session_tsp(name, budget, seed = 0)

Arguments

Details

UNLIKE every other sz_eval_session_*() constructor, this session's $evaluate() rows and $random_permutation()/$best() tour values are all 1-BASED (a permutation of 1:n_cities) – see sz_tsp_tour_length's own doc, "Index-convention decision" – NOT the dim-length numeric coordinates every continuous session above expects. The returned object additionally exposes:

• $kind() – always "permutation" for a session built by this constructor.

• $random_permutation() – a uniformly random 1-based tour (a permutation of 1:dim()), drawn from this session's own seeded Rust-side RNG stream (see "RNG parity with py-sezgi" below).

$dim() returns the instance's n_cities; $bounds() errors (there is no uniform (lo, hi) domain for a tour – use $random_permutation() instead of ctx$random_point() when authoring an algorithm over this session via sz_algorithm). Every other method ($evaluate(), $evals_used(), $budget(), $best(), $f_opt(), $finish()) behaves the same as for a continuous session, just over 1-based tour rows instead of coordinate rows.

RNG parity with py-sezgi: $random_permutation() draws from a Rust-side RngStream seeded via RngStream::from_master(seed, &[PERM_SESSION_RNG_TAG]) – the IDENTICAL tag value and derivation py-sezgi's own PermSession uses, over the SAME shared sezgi_components::perm::fisher_yates_shuffle core (see src/rust/src/session.rs's module doc, "Permutation-typed sessions"). So for the SAME seed, an R session built here and a Python session built via sezgi.EvalSession.for_problem(sezgi.problems.tsp(name)) draw the bit-identical UNDERLYING 0-based permutation on their first random_permutation() call – R simply displays it shifted +1.

Deliberately has NO log_dir/algo_name arguments: IOH logging is not wired up for permutation-typed sessions (no suite/fid identity exists to log a TSP run against) – same choice sz_eval_session_f0 already makes for its own unsupported-logging case.

Value

An 'EvalSession' object (see above and sz_eval_session).

A continuous block: 'n' coordinates, each in '[lo, hi]'.

Description

Mirrors 'sezgi.Float' ('py-sezgi/python/sezgi/spaces.py') exactly: a plain value object, fields stored verbatim with no construction-time validation (bounds are enforced only once the space reaches the Rust core, e.g. 'SearchSpace::new', 'crates/core/src/space.rs').

Usage

sz_float(lo, hi, n)

Arguments

Value

An object of class 'c("sz_block_float", "sz_block")' with elements 'lo', 'hi', 'n'.

Examples

sz_float(-5, 5, 3)

An integer block: 'n' coordinates, each in '[lo, hi]' (inclusive).

Description

Mirrors 'sezgi.Int' ('py-sezgi/python/sezgi/spaces.py') exactly – see [sz_float()]'s doc for the no-construction-time-validation note.

Usage

sz_int(lo, hi, n)

Arguments

Value

An object of class 'c("sz_block_int", "sz_block")' with elements 'lo', 'hi', 'n'.

Examples

sz_int(0L, 10L, 2)

Direct, one-shot objective evaluation of a decision vector 'x' against any ‘sz_mo_*' problem string, bypassing 'sz_nsga2()'’s population/budget machinery entirely. Added (M3-7) so fixture-value tests can pin an exact ‘x' ('sz_nsga2()'’s randomly-initialized population cannot), mirroring the existing 'sz_cec2022_evaluate()'/'sz_cec2014_evaluate()'/ 'sz_cec2017_evaluate()' one-shot-evaluation convention.

Description

Direct, one-shot objective evaluation of a decision vector 'x' against any ‘sz_mo_*' problem string, bypassing 'sz_nsga2()'’s population/budget machinery entirely. Added (M3-7) so fixture-value tests can pin an exact ‘x' ('sz_nsga2()'’s randomly-initialized population cannot), mirroring the existing 'sz_cec2022_evaluate()'/'sz_cec2014_evaluate()'/ 'sz_cec2017_evaluate()' one-shot-evaluation convention.

Usage

sz_mo_evaluate(problem, dim, x, m = NULL, k = NULL, l = NULL)

Arguments

Value

A numeric vector, length equal to the problem's own objective count.

The matching one-shot constraint-row evaluation for 'sz_mo_evaluate()', same calling convention.

Description

The matching one-shot constraint-row evaluation for 'sz_mo_evaluate()', same calling convention.

Usage

sz_mo_evaluate_constraints(problem, dim, x, m = NULL, k = NULL, l = NULL)

Arguments

Value

A numeric vector ('g_1..g_ncon', 'g_j >= 0' meaning SATISFIED), or 'NULL' for an unconstrained problem (every zdt/wfg problem, and dtlz1-7).

General-M exact hypervolume (While, Bradstreet & Barone 2012, the WFG algorithm; 'M == 2' delegates internally to the SAME ['sezgi_stats::hypervolume_2d']) – binds ['sezgi_stats::hypervolume'] (M3-7 Task 8/11). Unlike 'sz_mo_hypervolume_2d', 'front'/'ref_point' may have any number 'M >= 1' of objectives.

Description

‘ref_point' is REQUIRED, with no default: 'sezgi_stats::moo_indicators'’s own module doc, "Choosing a reference point: explicit-always, contested in the literature" section, deliberately never picks one for the caller. One common convention from that literature (also critiqued by Ishibuchi, Imada, Setoguchi & Nojima 2018, "How to Specify a Reference Point in Hypervolume Calculation for Fair Performance Comparison", GECCO Companion) is the analytic front's nadir point (the componentwise worst value across the front) scaled by '1.1' – a caller-supplied choice, never defaulted here.

Usage

sz_mo_hypervolume(front, ref_point)

Arguments

Value

A numeric scalar. An EMPTY 'front' is NOT an error – it returns ‘0.0' (the algorithm’s own base case).

# Errors A savvy error for any ['sezgi_stats::StatsError'] ('ref_point' empty, a 'front' row with a different number of objectives than 'ref_point', or a non-finite value), or if 'front' is not a matrix.

Exact 2-objective hypervolume (Zitzler & Thiele 1999 S-metric, reference-point variant; minimization) – binds [‘sezgi_stats::hypervolume_2d'] exactly. See that function’s doc for the pinned definition.

Description

Exact 2-objective hypervolume (Zitzler & Thiele 1999 S-metric, reference-point variant; minimization) – binds [‘sezgi_stats::hypervolume_2d'] exactly. See that function’s doc for the pinned definition.

Usage

sz_mo_hypervolume_2d(front, ref_point)

Arguments

Value

A numeric scalar.

# Errors A savvy error if 'ref_point' does not have exactly 2 values, if 'front' is not a matrix, or for any ['sezgi_stats::StatsError'] (empty front, a non-2-objective row, or a non-finite value).

Inverted Generational Distance (Ishibuchi et al. 2015, eq. 12, 'p = 1') – binds ['sezgi_stats::igd'] exactly. Any (equal, consistent) number of objectives across both 'front' and 'reference_front'.

Description

Inverted Generational Distance (Ishibuchi et al. 2015, eq. 12, 'p = 1') – binds ['sezgi_stats::igd'] exactly. Any (equal, consistent) number of objectives across both 'front' and 'reference_front'.

Usage

sz_mo_igd(front, reference_front)

Arguments

Value

A numeric scalar.

# Errors A savvy error if either argument is not a matrix, or for any ['sezgi_stats::StatsError'] (an empty 'front'/'reference_front', a dimension mismatch, or a non-finite value).

A deterministic 'n'-point sample of the analytic Pareto front in OBJECTIVE space, if known. See 'MoProblem::pareto_front' in 'crates/core/src/mo.rs'.

Description

A deterministic 'n'-point sample of the analytic Pareto front in OBJECTIVE space, if known. See 'MoProblem::pareto_front' in 'crates/core/src/mo.rs'.

Usage

sz_mo_pareto_front(problem, dim, n, m = NULL, k = NULL, l = NULL)

Arguments

Value

A ‘list' of 'n' numeric vectors (mirrors 'sz_nsga2()'’s 'objectives'/'individuals' container convention – a list, not a matrix), or 'NULL' when the problem has no known analytic front sample at this 'm' (verified cases: DTLZ5/DTLZ6 with 'm > 3'; WFG1/WFG2 unconditionally) – mirrors py-sezgi's 'sezgi.mo.pareto_front()' return exactly.

Reads a "sezgi-moa v1" archive file written by 'sz_nsga2(..., log_dir =, label =)' (M3-7 Task 9/11) – binds ['sezgi_bench::read_moa']. Returns a named list: - 'algo' (character): the logging algorithm name – always '"nsga2"' today (‘nsga2_run_logged'’s own fixed 'NSGA2_ALGO_NAME'). - 'problem' (character): the 'label' 'sz_nsga2' was called with. **Kept as the literal on-disk header key name** (‘crates/bench/src/mo_archive.rs'’s own format grammar: the header line is 'problem <label>', not 'label <label>') rather than renamed here to '"label"' – this binding stays a thin, direct mirror of ‘MoArchiveRun'’s own field names, so a reader cross-checking against the Rust struct (or the Python binding, M3-7 Task 10) sees the SAME key everywhere. - 'm', 'seed', 'budget' (double, whole-number-valued – R has no native integer64). - 'kind' (character): '"float"' or '"binary"'. - 'records' (list, in file/eval order): each entry a named list with 'eval_index' (double), 'objectives' (numeric vector), 'genotype' (numeric vector for 'kind = "float"', 'logical' vector for ‘kind = "binary"' – see this module’s own doc, "Container-idiom decisions"). - 'archive' (list of numeric vectors): the reconstructed nondominated archive at evaluation budget 'at', via 'MoArchiveRun::archive_at'.

Description

‘# sezgi decision:' 'at = NULL' (the default) resolves to the file’s own logged ‘budget' header field (the full run’s final archive) – mirrors py-sezgi's own 'mo.read_moa(path, at=None)' default exactly.

Usage

sz_mo_read_moa(path, at = NULL)

Arguments

Details

This is directly '@export'ed (no hand-written wrapper needed): 'at' is the only optional parameter and trails 'path', so savvy already emits 'at = NULL' in the generated R signature – same pattern as 'sz_mo_hypervolume_2d'/'sz_mo_igd'.

Value

A named list – see this function's own doc above.

# Errors A savvy error for any ['sezgi_bench::MoArchiveError'] (missing file, a malformed header, or a malformed record line).

NSGA-II (Deb, Pratap, Agarwal & Meyarivan 2002) run on a ZDT/DTLZ/WFG multi-objective test problem. See 'crates/components/src/nsga2.rs' for the full algorithm provenance.

Description

NSGA-II (Deb, Pratap, Agarwal & Meyarivan 2002) run on a ZDT/DTLZ/WFG multi-objective test problem. See 'crates/components/src/nsga2.rs' for the full algorithm provenance.

Usage

sz_nsga2(
  problem,
  dim,
  m = NULL,
  pop_size,
  budget,
  seed = 0,
  eta_c = 20,
  eta_m = 20,
  p_c = 0.9,
  p_m = NULL,
  p_c_bin = 0.9,
  p_m_bin = NULL,
  p_c_cat = 0.9,
  p_m_cat = NULL,
  k = NULL,
  l = NULL,
  log_dir = NULL,
  label = NULL
)

Arguments

Value

A named list with 'individuals' (list of numeric vectors, one per final-population member, each length 'dim' – flattened across every block; a ‘Block::Binary' block’s bits are flattened to ‘0.0'/'1.0', e.g. zdt5’s own path), 'objectives' (list of numeric vectors, parallel to ‘individuals', each length equal to the problem’s objective count), 'front0' (numeric vector of 1-based positions, R convention, into ‘individuals'/'objectives' for the final population’s non-dominated set – NOTE: py-sezgi's 'front0' is 0-based; this binding converts), 'evals_used' (double), and (present ONLY for a constrained problem – dtlz8/dtlz9 today) 'violations' (numeric vector, '<= 0.0', '0.0' = feasible, parallel to 'individuals'/'objectives') – mirrors py-sezgi's 'sezgi.mo.nsga2()' dict keys exactly.

Examples

r <- sz_nsga2("zdt1", dim = 5, pop_size = 8, budget = 40, seed = 1)
length(r$front0)

Build one paper-package statistics list PER DISTINCT BUDGET present in a 'sz_run_experiment()' data.frame, in ascending budget order.

Description

Mirrors py-sezgi's 'per_budget_packages()'. Piotrowski et al. (2025) show algorithm rankings on benchmark comparisons can flip depending on which evaluation budget is examined, so this makes multi-budget reporting the default rather than a single, arbitrarily-chosen budget's report: compare algorithms per budget, never pooled across budgets.

Usage

sz_per_budget_packages(
  df,
  rope = 0,
  samples = 20000,
  seed = 1,
  aggregate = "mean"
)

Arguments

Value

A named list, one entry per distinct budget in ascending order, named by the budget (as a string); each value has exactly the shape 'sz_stats_paper_package()' returns.

A permutation block: one permutation of '0..n'.

Description

Mirrors 'sezgi.Permutation' ('py-sezgi/python/sezgi/spaces.py') exactly – see [sz_float()]'s doc for the no-construction-time-validation note.

Usage

sz_permutation(n)

Arguments

Value

An object of class 'c("sz_block_permutation", "sz_block")' with element 'n'.

Builds an Artificial Bee Colony spec (Karaboga 2005, TR-06 / Karaboga & Basturk 2007, Journal of Global Optimization – a labeled metaphor preset; see ‘crates/components/src/abc.rs'’s module doc for the full provenance extraction against the author's own 'ABCorig.m' plus the official 'Python_ABC' port, the pop<->food-source convention resolution, the fitness-transform monotonicity proof, and the phase-design adjudication – a SINGLE stage, not two symmetric stages like 'tlbo': 'gen/abc-employed' + the new 'replace/abc-trial-greedy', plus the new 'adapter/abc-onlooker-scout' folding the onlooker AND scout phases together) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds an Artificial Bee Colony spec (Karaboga 2005, TR-06 / Karaboga & Basturk 2007, Journal of Global Optimization – a labeled metaphor preset; see ‘crates/components/src/abc.rs'’s module doc for the full provenance extraction against the author's own 'ABCorig.m' plus the official 'Python_ABC' port, the pop<->food-source convention resolution, the fitness-transform monotonicity proof, and the phase-design adjudication – a SINGLE stage, not two symmetric stages like 'tlbo': 'gen/abc-employed' + the new 'replace/abc-trial-greedy', plus the new 'adapter/abc-onlooker-scout' folding the onlooker AND scout phases together) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_abc(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds an Ant Lion Optimizer spec (Mirjalili 2015, Advances in Engineering Software – a labeled metaphor preset; see ‘crates/components/src/alo.rs'’s module doc for the full provenance extraction against the author's own 'ALO.m'/'Random_walk_around_ antlion.m'/'RouletteWheelSelection.m', the faithful-full-walk cost decision, and the elitism design adjudication – the antlion population itself is the persisted memory via 'replace/mu-plus-lambda', no blackboard state needed) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds an Ant Lion Optimizer spec (Mirjalili 2015, Advances in Engineering Software – a labeled metaphor preset; see ‘crates/components/src/alo.rs'’s module doc for the full provenance extraction against the author's own 'ALO.m'/'Random_walk_around_ antlion.m'/'RouletteWheelSelection.m', the faithful-full-walk cost decision, and the elitism design adjudication – the antlion population itself is the persisted memory via 'replace/mu-plus-lambda', no blackboard state needed) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_alo(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Bat Algorithm spec (Yang, X.-S. 2010, NICSO – a labeled metaphor preset, see ‘crates/components/src/ba.rs'’s module doc for the tier note, citation, the verified 'bat_algorithm.m' loop structure, the verified fixed-loudness/pulse-rate finding, the two composing sign-inversion deltas in the frequency draw and velocity term, and the design adjudication for the new 'replace/bat-loudness-greedy' acceptance-coupled replacer) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Bat Algorithm spec (Yang, X.-S. 2010, NICSO – a labeled metaphor preset, see ‘crates/components/src/ba.rs'’s module doc for the tier note, citation, the verified 'bat_algorithm.m' loop structure, the verified fixed-loudness/pulse-rate finding, the two composing sign-inversion deltas in the frequency draw and velocity term, and the design adjudication for the new 'replace/bat-loudness-greedy' acceptance-coupled replacer) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_bat(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Table-driven preset-backed algorithm wrapper classes.

Description

One R6 class per non-GA/DE 'sz_preset_*' builder ('R/000-wrappers.R' + 'R/presets.R', both mirroring 'crates/components/src/presets.rs') – 26 classes, generated from ONE row table ('.sz_preset_table') by ONE factory ('.sz_make_preset_class()', both internal to 'R/builtins.R'). Every class shares the identical shape: '$new(pop_size = NULL, ...)' stores constructor kwargs only (no computation happens at construction time); '$run(problem, budget, seed = 0, run_id = 0)' builds the preset's 'spec_json' via the matching 'sz_preset_*(...)' call (bit-identical to calling that function directly with the same arguments – see ‘test-oop-builtins.R'’s own table-driven anchor test), routes it through 'sz_solve_bbob()'/'sz_solve_r_problem()' depending on ‘problem'’s own form, and wraps the raw 3-field result via T4's ‘.sz_wrap_result()' into an 'sz_result' (the SAME shape [Algorithm]’s own 'run()' returns).

Details

‘pop_size' semantics depend on the preset’s own Rust signature (see each ‘sz_preset_*()'’s own doc in 'R/000-wrappers.R' for its "Canonical is N" convention, where stated):

most presets

'sz_preset_*(pop_size, budget, ...)' – ‘pop_size' defaults to the class’s own documented convention (the source paper's canonical value where '000-wrappers.R' states one; ‘20' otherwise, matching this repo’s existing dominant convention for presets with no single canonical source), and MAY be overridden at '$new()'.

'SimulatedAnnealing'

'sz_preset_sa(budget)' has NO 'pop_size' parameter at all (a single search trajectory) – 'pop_size' is fixed at '1'; a non-'NULL', non-'1' value at '$new()' raises a clear error.

'CMAESIpop'/'LSHADE'/'NelderMead'

'sz_preset_*(dim, budget, ...)‘ DERIVES its own population size from the problem’s dimensionality at '$run()' time ('dim' is read from the resolved problem, NOT 'pop_size') – ANY 'pop_size' given at '$new()' raises a clear error naming the derivation formula.

'EvolutionStrategy' additionally accepts 'dist='/'mean='/'sigma='/ 'loc='/'scale='/'alpha='/'nu=' at '$new()' (flowing through to 'sz_preset_es_mu_plus_lambda()' unchanged, via '...') – the ONLY table row whose preset takes kwargs beyond 'pop_size'/'budget'.

Problem-form support matrix (identical for every class in this family, and for [GeneticAlgorithm]/[DifferentialEvolution]): 'problem' is EITHER a [Problem] subclass instance (routed through 'sz_solve_r_problem()', 'f_opt = prob$optimum()') OR an [sz_builtin_bbob()] descriptor (routed through 'sz_solve_bbob()', ‘f_opt = NULL') – the SAME two forms [Algorithm]’s own 'run()' accepts. No CEC2014/CEC2017/CEC2022/TSP/onemax/etc. native-descriptor path exists yet.

Builds a (mu/mu_w,lambda)-CMA-ES algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a (mu/mu_w,lambda)-CMA-ES algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_cmaes(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a CMA-ES with IPOP-style stagnation restarts algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a CMA-ES with IPOP-style stagnation restarts algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_cmaes_ipop(dim, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Cuckoo Search algorithm spec (Yang & Deb 2009 – a labeled metaphor preset, see ‘crates/components/src/cs.rs'’s module doc for the tier note, citation and pinned draw order) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Cuckoo Search algorithm spec (Yang & Deb 2009 – a labeled metaphor preset, see ‘crates/components/src/cs.rs'’s module doc for the tier note, citation and pinned draw order) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_cuckoo_search(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a DE/best/1/bin algorithm spec (uniform init, clamp boundary, one-to-one-greedy replacement) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a DE/best/1/bin algorithm spec (uniform init, clamp boundary, one-to-one-greedy replacement) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_de_best_1(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a DE/rand/1/bin algorithm spec (uniform init, clamp boundary, one-to-one-greedy replacement) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a DE/rand/1/bin algorithm spec (uniform init, clamp boundary, one-to-one-greedy replacement) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_de_rand_1(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Examples

sz_preset_de_rand_1(pop_size = 10, budget = 100)

Builds a (mu/mu_w,lambda)-ES algorithm spec (mutation step drawn from 'dist') as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Mirrors py-sezgi's 'sezgi.presets.es_mu_plus_lambda()'.

Usage

sz_preset_es_mu_plus_lambda(
  pop_size,
  budget,
  dist = "gaussian",
  mean = 0,
  sigma = 0.5,
  loc = 0,
  scale = 1,
  alpha = 1.5,
  nu = 3
)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Firefly Algorithm (Yang, X.-S., *Nature-Inspired Metaheuristic Algorithms*, 2nd ed., Luniver Press, 2010 – a labeled metaphor preset, see ‘crates/components/src/fa.rs'’s module doc for the tier note, citation, the verified 'fa_ndim.m'/'ffa_move.m' loop structure, the floored attractiveness formula, the closed-form 'alpha' decay, and the hybrid in-place-self/live-distance/frozen-target double-loop semantics) spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Firefly Algorithm (Yang, X.-S., *Nature-Inspired Metaheuristic Algorithms*, 2nd ed., Luniver Press, 2010 – a labeled metaphor preset, see ‘crates/components/src/fa.rs'’s module doc for the tier note, citation, the verified 'fa_ndim.m'/'ffa_move.m' loop structure, the floored attractiveness formula, the closed-form 'alpha' decay, and the hybrid in-place-self/live-distance/frozen-target double-loop semantics) spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_firefly(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Flower Pollination Algorithm spec (Yang, X.-S. 2012, UCNC – a labeled metaphor preset, see ‘crates/components/src/fpa.rs'’s module doc for the tier note, citation, the verified 'fpa_demo.m' loop structure, the switch-branch orientation delta, the global-step sign delta reusing ‘cs.rs'’s 'cs_dim_step' verbatim, the local-step self-selection-not- excluded finding, and the min_pop adjustment from 3 down to 2) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Flower Pollination Algorithm spec (Yang, X.-S. 2012, UCNC – a labeled metaphor preset, see ‘crates/components/src/fpa.rs'’s module doc for the tier note, citation, the verified 'fpa_demo.m' loop structure, the switch-branch orientation delta, the global-step sign delta reusing ‘cs.rs'’s 'cs_dim_step' verbatim, the local-step self-selection-not- excluded finding, and the min_pop adjustment from 3 down to 2) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_fpa(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Binary-space GA spec (tournament selection, uniform crossover, bit-flip mutation – 'gen/ga-bin' + 'replace/mu-plus-lambda') as JSON, ready to pass to 'sz_solve_onemax()'. Binds ['sezgi_components::presets::ga_bin'] exactly – M3-8 Task 10, mirroring py-sezgi's 'sezgi.presets.ga_bin' (M3-8 Task 9).

Description

Builds a Binary-space GA spec (tournament selection, uniform crossover, bit-flip mutation – 'gen/ga-bin' + 'replace/mu-plus-lambda') as JSON, ready to pass to 'sz_solve_onemax()'. Binds ['sezgi_components::presets::ga_bin'] exactly – M3-8 Task 10, mirroring py-sezgi's 'sezgi.presets.ga_bin' (M3-8 Task 9).

Usage

sz_preset_ga_bin(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Categorical-space GA spec (tournament selection, uniform crossover, random-reset mutation – 'gen/ga-cat' + 'replace/mu-plus-lambda') as JSON, ready to pass to 'sz_solve_cat_match()'. Binds ['sezgi_components::presets::ga_cat'] exactly – M3-8 Task 10, mirroring py-sezgi's 'sezgi.presets.ga_cat' (M3-8 Task 9).

Description

Builds a Categorical-space GA spec (tournament selection, uniform crossover, random-reset mutation – 'gen/ga-cat' + 'replace/mu-plus-lambda') as JSON, ready to pass to 'sz_solve_cat_match()'. Binds ['sezgi_components::presets::ga_cat'] exactly – M3-8 Task 10, mirroring py-sezgi's 'sezgi.presets.ga_cat' (M3-8 Task 9).

Usage

sz_preset_ga_cat(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds an Int-space GA spec (tournament selection, SBX-style integer crossover, polynomial-style integer mutation – 'gen/ga-int' + 'replace/mu-plus-lambda') as JSON, ready to pass to 'sz_solve_int_quadratic()'. Binds ['sezgi_components::presets::ga_int'] exactly – M3-8 Task 10, mirroring py-sezgi's 'sezgi.presets.ga_int' (M3-8 Task 9).

Description

Builds an Int-space GA spec (tournament selection, SBX-style integer crossover, polynomial-style integer mutation – 'gen/ga-int' + 'replace/mu-plus-lambda') as JSON, ready to pass to 'sz_solve_int_quadratic()'. Binds ['sezgi_components::presets::ga_int'] exactly – M3-8 Task 10, mirroring py-sezgi's 'sezgi.presets.ga_int' (M3-8 Task 9).

Usage

sz_preset_ga_int(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a permutation-space GA spec (tournament selection, order crossover, swap mutation – 'gen/ga-perm' + 'replace/mu-plus-lambda') as JSON, ready to pass to 'sz_solve_tsp()'. Binds [‘sezgi_components::presets::ga_perm'] exactly – same preset py-sezgi’s 'sezgi.presets.ga_perm' binds (M3-3 Task 9).

Description

Builds a permutation-space GA spec (tournament selection, order crossover, swap mutation – 'gen/ga-perm' + 'replace/mu-plus-lambda') as JSON, ready to pass to 'sz_solve_tsp()'. Binds [‘sezgi_components::presets::ga_perm'] exactly – same preset py-sezgi’s 'sezgi.presets.ga_perm' binds (M3-3 Task 9).

Usage

sz_preset_ga_perm(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a real-coded GA (SBX crossover, polynomial mutation) algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a real-coded GA (SBX crossover, polynomial mutation) algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_ga_real(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Grasshopper Optimisation Algorithm spec (Saremi, Mirjalili & Lewis 2017 – a labeled metaphor preset, see ‘crates/components/src/goa.rs'’s module doc for the tier note, citation, the IMPLEMENTER-VERIFY distance-normalization resolution and the zero-RNG-draw arithmetic-order pin) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Grasshopper Optimisation Algorithm spec (Saremi, Mirjalili & Lewis 2017 – a labeled metaphor preset, see ‘crates/components/src/goa.rs'’s module doc for the tier note, citation, the IMPLEMENTER-VERIFY distance-normalization resolution and the zero-RNG-draw arithmetic-order pin) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_goa(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Gravitational Search Algorithm spec (Rashedi, Nezamabadi-pour & Saryazdi 2009, Information Sciences – a labeled metaphor preset, and the wave's LAST stateful/blackboard algorithm; see ‘crates/components/src/gsa.rs'’s module doc for the full provenance extraction against the author's own 'GSA.m'/'Gconstant.m'/ 'massCalculation.m'/'Gfield.m'/'move.m', the verified 'M_i'-free force delta, and the confirmation that GSA's own 'Fbest'/'Lbest' never feed back into the mechanism) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Gravitational Search Algorithm spec (Rashedi, Nezamabadi-pour & Saryazdi 2009, Information Sciences – a labeled metaphor preset, and the wave's LAST stateful/blackboard algorithm; see ‘crates/components/src/gsa.rs'’s module doc for the full provenance extraction against the author's own 'GSA.m'/'Gconstant.m'/ 'massCalculation.m'/'Gfield.m'/'move.m', the verified 'M_i'-free force delta, and the confirmation that GSA's own 'Fbest'/'Lbest' never feed back into the mechanism) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_gsa(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Grey Wolf Optimizer algorithm spec (Mirjalili, Mirjalili & Lewis 2014 – a labeled metaphor preset, see ‘crates/components/src/gwo.rs'’s module doc for the tier note and citations) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Grey Wolf Optimizer algorithm spec (Mirjalili, Mirjalili & Lewis 2014 – a labeled metaphor preset, see ‘crates/components/src/gwo.rs'’s module doc for the tier note and citations) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_gwo(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Harmony Search algorithm spec (Geem, Kim & Loganathan 2001 – a labeled metaphor preset, see ‘crates/components/src/hs.rs'’s module doc for the tier note and citations) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Harmony Search algorithm spec (Geem, Kim & Loganathan 2001 – a labeled metaphor preset, see ‘crates/components/src/hs.rs'’s module doc for the tier note and citations) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_harmony_search(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Harris Hawks Optimization spec (Heidari, Mirjalili, Faris, Aljarah, Mafarja & Chen 2019, Future Generation Computer Systems – a labeled metaphor preset, and the wave's most structurally complex one: a multi-branch escape-energy tree whose progressive rapid-dive sub-branches evaluate mid-'generate()'. See ‘crates/components/src/hho.rs'’s module doc for the full provenance extraction against the paper author's own 'HHO.m', the hard/soft besiege mapping delta, the mean(X)/random-hawk in-place semantics, and the prominent in-generator-evaluation eval-accounting design decision) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Harris Hawks Optimization spec (Heidari, Mirjalili, Faris, Aljarah, Mafarja & Chen 2019, Future Generation Computer Systems – a labeled metaphor preset, and the wave's most structurally complex one: a multi-branch escape-energy tree whose progressive rapid-dive sub-branches evaluate mid-'generate()'. See ‘crates/components/src/hho.rs'’s module doc for the full provenance extraction against the paper author's own 'HHO.m', the hard/soft besiege mapping delta, the mean(X)/random-hawk in-place semantics, and the prominent in-generator-evaluation eval-accounting design decision) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_hho(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a JAYA spec (Rao 2016 – a labeled metaphor preset, see ‘crates/components/src/jaya.rs'’s module doc for the tier note, citation, the primary-paper-verified worked-example reproduction, the shared-per-dimension-per-generation 'r1'/'r2' draw finding and the greedy-replacement delta vs mealpy's misleadingly-named 'OriginalJA') as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a JAYA spec (Rao 2016 – a labeled metaphor preset, see ‘crates/components/src/jaya.rs'’s module doc for the tier note, citation, the primary-paper-verified worked-example reproduction, the shared-per-dimension-per-generation 'r1'/'r2' draw finding and the greedy-replacement delta vs mealpy's misleadingly-named 'OriginalJA') as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_jaya(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a jDE algorithm spec (self-adaptive F/CR DE) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a jDE algorithm spec (self-adaptive F/CR DE) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_jde(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds an L-SHADE algorithm spec (population linearly reduced from '18 * dim') as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds an L-SHADE algorithm spec (population linearly reduced from '18 * dim') as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_lshade(dim, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds an MFO (Moth-Flame Optimization; Mirjalili 2015 – a labeled metaphor preset, see ‘crates/components/src/mfo.rs'’s module doc for the tier note, citation, the verified 'MFO.m' loop structure, the two subtle draw/index deltas found vs the plan's sketch, and the blackboard flame-memory design) spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds an MFO (Moth-Flame Optimization; Mirjalili 2015 – a labeled metaphor preset, see ‘crates/components/src/mfo.rs'’s module doc for the tier note, citation, the verified 'MFO.m' loop structure, the two subtle draw/index deltas found vs the plan's sketch, and the blackboard flame-memory design) spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_mfo(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Nelder-Mead simplex algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Nelder-Mead simplex algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_nelder_mead(dim, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a PSO (Clerc-Kennedy constriction) algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a PSO (Clerc-Kennedy constriction) algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_pso(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a random search algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a random search algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_random_search(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a simulated annealing (Metropolis, geometric cooling) algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a simulated annealing (Metropolis, geometric cooling) algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_sa(budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Sine Cosine Algorithm spec (Mirjalili 2016 – a labeled metaphor preset, see ‘crates/components/src/sca.rs'’s module doc for the tier note, citation, the 'SCA.m'-verified pinned draw order and the mealpy-'OriginalSCA' replacer delta) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Sine Cosine Algorithm spec (Mirjalili 2016 – a labeled metaphor preset, see ‘crates/components/src/sca.rs'’s module doc for the tier note, citation, the 'SCA.m'-verified pinned draw order and the mealpy-'OriginalSCA' replacer delta) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_sca(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a SHADE algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a SHADE algorithm spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_shade(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds an SSA (Salp Swarm Algorithm; Mirjalili et al. 2017 – a labeled metaphor preset, see ‘crates/components/src/ssa.rs'’s module doc for the tier note, citation, the verified 'SSA.m' half-population leader/follower split, the leader sign-branch pin, the verified in-place follower-chain semantics, and the persisted-food-vs-current-pop-best delta) spec as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds an SSA (Salp Swarm Algorithm; Mirjalili et al. 2017 – a labeled metaphor preset, see ‘crates/components/src/ssa.rs'’s module doc for the tier note, citation, the verified 'SSA.m' half-population leader/follower split, the leader sign-branch pin, the verified in-place follower-chain semantics, and the persisted-food-vs-current-pop-best delta) spec as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_ssa(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Teaching-Learning-Based Optimization spec (Rao, Savsani & Vakharia 2011, Computer-Aided Design – a labeled metaphor preset, and sezgi's FIRST multi-stage preset: two '[[stages]]' (teacher, then learner) run in sequence every generation. See ‘crates/components/src/tlbo.rs'’s module doc for the full provenance extraction against Yarpiz's 'tlbo.m' – explicitly labeled third-party, not Rao's own code – the per-learner teaching-factor finding, the unconditionally-distinct partner-selection finding, the min_pop adjustment from 3 down to 2, and the "parameter-free" framing's Črepinšek/Liu/Mernik (2012) counterpoint) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Teaching-Learning-Based Optimization spec (Rao, Savsani & Vakharia 2011, Computer-Aided Design – a labeled metaphor preset, and sezgi's FIRST multi-stage preset: two '[[stages]]' (teacher, then learner) run in sequence every generation. See ‘crates/components/src/tlbo.rs'’s module doc for the full provenance extraction against Yarpiz's 'tlbo.m' – explicitly labeled third-party, not Rao's own code – the per-learner teaching-factor finding, the unconditionally-distinct partner-selection finding, the min_pop adjustment from 3 down to 2, and the "parameter-free" framing's Črepinšek/Liu/Mernik (2012) counterpoint) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_tlbo(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Builds a Whale Optimization Algorithm spec (Mirjalili & Lewis 2016 – a labeled metaphor preset, see ‘crates/components/src/woa.rs'’s module doc for the tier note and citations) as JSON, ready to pass to 'sz_solve_bbob()'.

Description

Builds a Whale Optimization Algorithm spec (Mirjalili & Lewis 2016 – a labeled metaphor preset, see ‘crates/components/src/woa.rs'’s module doc for the tier note and citations) as JSON, ready to pass to 'sz_solve_bbob()'.

Usage

sz_preset_woa(pop_size, budget)

Arguments

Value

A character scalar with the algorithm spec as JSON.

Reconstructs 'RunRecord's from an on-disk IOH archive at 'log_root' (as written by 'sz_run_experiment(..., log_dir = ...)'), one record per ‘(run, budget)' pair – see 'sezgi_bench::ioh_records'’s doc comment for the exact 'best_f'/'evals_used' semantics and the curtailed-view-vs- independent-run distinction for budgets smaller than a run's logged budget.

Description

Returns the SAME data.frame shape 'sz_run_experiment()' returns (via ['records_to_data_frame']), so 'sz_results_matrix()'/ 'sz_per_budget_packages()' accept it unchanged.

Usage

sz_read_ioh_records(log_root, budgets)

Arguments

Value

A data.frame with the same columns as 'sz_run_experiment()'.

Build a 'sezgi_stats'-shaped results matrix for one 'budget' from a 'sz_run_experiment()' data.frame.

Description

Mirrors py-sezgi's 'results_matrix()'.

Usage

sz_results_matrix(df, budget, aggregate = "mean")

Arguments

Value

A named list with 'algo_names' (character vector), 'problem_labels' (character vector, 'f{fid}d{dim}i{instance}' for BBOB rows, '{short}-f{fid}d{dim}i{instance}' for any other suite, ordered by first appearance in 'df'), and 'matrix' (numeric matrix, rows = problems, columns = algorithms; 'matrix[i, j]' is the aggregated gap of 'algo_names[j]' on 'problem_labels[i]'). Errors if a (problem, algorithm) pair present for one algorithm/problem is missing for another at this budget (an incomplete experiment).

Run a sezgi experiment spec across algorithms x problems x instances x seeds x budgets.

Description

Mirrors py-sezgi's 'run_experiment()': 'journal = NULL' runs directly (in parallel or sequentially, per 'parallel'); a non-'NULL' 'journal' path runs through the checkpoint executor, which resumes from – and appends to – an existing journal file at that path.

Usage

sz_run_experiment(
  spec_toml,
  journal = NULL,
  parallel = TRUE,
  threads = NULL,
  log_dir = NULL
)

Arguments

Value

A data.frame with one row per run and columns 'algo', 'fid', 'dim', 'instance', 'seed', 'budget', 'suite', 'best_f', 'f_opt', 'evals'.

Examples

spec_toml <- '
name = "quick"
seeds = [1]
budgets = [100]

[[algorithms]]
name = "de1"
preset = { kind = "de_rand_1", pop_size = 10 }

[[problems]]
suite = "bbob"
fid = 1
dim = 2
instances = [1]
'
df <- sz_run_experiment(spec_toml)
df$evals

Runs an algorithm spec on a BBOB problem and returns the result.

Description

Runs an algorithm spec on a BBOB problem and returns the result.

Usage

sz_solve_bbob(spec_json, fid, dim, instance, master_seed, run_id)

Arguments

Value

A named list with 'best_f' (double), 'evals' (double), and 'best_x' (double vector) – the best EVALUATED point (paired with 'best_f'). For most algorithms this always lies within the declared domain. It is not guaranteed to for every algorithm: some (e.g. HHO) charge raw, pre-boundary-repair trial points against the budget before boundary repair, and such a point can become the reported best if it happens to be the run's own minimum.

Examples

spec <- sz_preset_de_rand_1(pop_size = 10, budget = 100)
r <- sz_solve_bbob(spec, fid = 1L, dim = 2L, instance = 1L, master_seed = 1, run_id = 0)
r$evals

Runs an algorithm spec on ['CatMatch'] (a Categorical-block Hamming- distance-to-target matching problem; 'sezgi_problems::diagnostics::CatMatch') and returns the result – M3-8 Task 10, mirroring 'sz_solve_onemax' exactly. Pairs with 'sz_preset_ga_cat(...)'. Diagnostic only, see ‘CatMatch'’s own module doc.

Description

Runs an algorithm spec on ['CatMatch'] (a Categorical-block Hamming- distance-to-target matching problem; 'sezgi_problems::diagnostics::CatMatch') and returns the result – M3-8 Task 10, mirroring 'sz_solve_onemax' exactly. Pairs with 'sz_preset_ga_cat(...)'. Diagnostic only, see ‘CatMatch'’s own module doc.

Usage

sz_solve_cat_match(spec_json, k, n, seed, master_seed, run_id)

Arguments

Value

A named list with 'best_f' (double), 'evals' (double), and 'best_x' (an INTEGER vector of category INDICES '0..k' – see ‘genotype_to_r'’s doc).

# Errors A savvy error for any ['sezgi_core::spec'] parse error, or any ['sezgi_core::engine'] run error.

Runs an algorithm spec on a CEC 2014 (Liang, Qu & Suganthan 2013) function via ['Cec2014::new'] and returns the result – M3-6 Task 10, mirroring 'sz_solve_cec2022' exactly ('Engine::from_spec' + 'engine.run' + result conversion): same 'best_f'/'evals'/'best_x' shape. See [‘Cec2014::new']’s own doc for the exact 'fid'/'dim' domain.

Description

Runs an algorithm spec on a CEC 2014 (Liang, Qu & Suganthan 2013) function via ['Cec2014::new'] and returns the result – M3-6 Task 10, mirroring 'sz_solve_cec2022' exactly ('Engine::from_spec' + 'engine.run' + result conversion): same 'best_f'/'evals'/'best_x' shape. See [‘Cec2014::new']’s own doc for the exact 'fid'/'dim' domain.

Usage

sz_solve_cec2014(spec_json, fid, dim, master_seed, run_id)

Arguments

Value

A named list with 'best_f' (double), 'evals' (double), and 'best_x' (double vector) – the best EVALUATED point (paired with ‘best_f'). Same caveat as 'sz_solve_bbob()': not every algorithm’s reported best is guaranteed to lie within the declared domain.

# Errors A savvy error for 'fid' outside '1..=30', 'dim' outside '{10,30}', any ['sezgi_core::spec'] parse error, or any ['sezgi_core::engine'] run error.

Runs an algorithm spec on a CEC 2017 (Awad, Ali, Liang, Qu & Suganthan 2016) function via ['Cec2017::new'] and returns the result – M3-6 Task 10, mirroring ‘sz_solve_cec2014' exactly. See ['Cec2017::new']’s own doc for the exact 'fid'/'dim' domain.

Description

Runs an algorithm spec on a CEC 2017 (Awad, Ali, Liang, Qu & Suganthan 2016) function via ['Cec2017::new'] and returns the result – M3-6 Task 10, mirroring ‘sz_solve_cec2014' exactly. See ['Cec2017::new']’s own doc for the exact 'fid'/'dim' domain.

Usage

sz_solve_cec2017(spec_json, fid, dim, master_seed, run_id)

Arguments

Value

A named list with 'best_f' (double), 'evals' (double), and 'best_x' (double vector) – the best EVALUATED point (paired with ‘best_f'). Same caveat as 'sz_solve_bbob()': not every algorithm’s reported best is guaranteed to lie within the declared domain.

# Errors A savvy error for 'fid' outside '{1} union {3..=30}', 'dim' outside '{10,30}', any ['sezgi_core::spec'] parse error, or any ['sezgi_core::engine'] run error. 'fid = 2' raises a dedicated error – the Rust ['sezgi_problems::Cec2017Error::Withdrawn'] message is surfaced VERBATIM.

Runs an algorithm spec on a CEC 2022 (Kumar, Price, Mohamed, Hadi & Suganthan 2021) function via ['Cec2022::new'] and returns the result – added later than the rest of the CEC surface, closing the gap where r-sezgi previously bound only direct evaluation ('sz_cec2022_evaluate'/'sz_cec2022_f_star'), with no 'solve()'-integrated path, unlike py-sezgi's 'sezgi.problems.cec2022(...)' + 'sezgi.solve()'. Mirrors 'sz_solve_bbob'/'sz_solve_tsp' exactly ('Engine::from_spec' + 'engine.run' + result conversion): same 'best_f'/'evals'/'best_x' shape, not py-sezgi's own 'solve()' dict shape ('best_f'/'best_x'/'evals_used'/ 'iterations') – the established r-sezgi 'sz_solve_*' convention governs here too. See ‘Cec2022::new'’s own doc for the exact 'fid'/'dim' domain.

Description

Runs an algorithm spec on a CEC 2022 (Kumar, Price, Mohamed, Hadi & Suganthan 2021) function via ['Cec2022::new'] and returns the result – added later than the rest of the CEC surface, closing the gap where r-sezgi previously bound only direct evaluation ('sz_cec2022_evaluate'/'sz_cec2022_f_star'), with no 'solve()'-integrated path, unlike py-sezgi's 'sezgi.problems.cec2022(...)' + 'sezgi.solve()'. Mirrors 'sz_solve_bbob'/'sz_solve_tsp' exactly ('Engine::from_spec' + 'engine.run' + result conversion): same 'best_f'/'evals'/'best_x' shape, not py-sezgi's own 'solve()' dict shape ('best_f'/'best_x'/'evals_used'/ 'iterations') – the established r-sezgi 'sz_solve_*' convention governs here too. See ‘Cec2022::new'’s own doc for the exact 'fid'/'dim' domain.

Usage

sz_solve_cec2022(spec_json, fid, dim, master_seed, run_id)

Arguments

Value

A named list with 'best_f' (double), 'evals' (double), and 'best_x' (double vector) – the best EVALUATED point (paired with ‘best_f'). Same caveat as 'sz_solve_bbob()': not every algorithm’s reported best is guaranteed to lie within the declared domain.

# Errors A savvy error for 'fid' outside '1..=12', 'dim' outside '{2,10,20}', 'dim = 2' for a hybrid function, any ['sezgi_core::spec'] parse error, or any ['sezgi_core::engine'] run error.

Runs an algorithm spec on ['IntQuadratic'] (an Int-block quadratic bowl around a fixed, deterministically-derived target; 'sezgi_problems::diagnostics::IntQuadratic') and returns the result – M3-8 Task 10, mirroring 'sz_solve_onemax' exactly. Pairs with ‘sz_preset_ga_int(...)'. Diagnostic only, see 'IntQuadratic'’s own module doc.

Description

Runs an algorithm spec on ['IntQuadratic'] (an Int-block quadratic bowl around a fixed, deterministically-derived target; 'sezgi_problems::diagnostics::IntQuadratic') and returns the result – M3-8 Task 10, mirroring 'sz_solve_onemax' exactly. Pairs with ‘sz_preset_ga_int(...)'. Diagnostic only, see 'IntQuadratic'’s own module doc.

Usage

sz_solve_int_quadratic(spec_json, lo, hi, n, master_seed, run_id)

Arguments

Value

A named list with 'best_f' (double), 'evals' (double), and ‘best_x' (an INTEGER vector – see 'genotype_to_r'’s doc).

# Errors A savvy error if 'lo >= hi', for any ['sezgi_core::spec'] parse error, or any ['sezgi_core::engine'] run error.

Runs an algorithm spec on [‘MixedDiagnostic'] (this file’s own Float+Int+Categorical+Binary mixed-space scaffold problem, see its own doc) and returns the result – M3-8 Task 10. Added SOLELY so 'gen/compound' (Task 5) is reachable end to end through the NORMAL R solve path, proven with a mixed-space 'AlgorithmSpec' authored as TOML (this task's own test) – UNLIKE its three siblings above (which take ‘spec_json', pairing with 'sz_preset_ga_bin/ga_int/ga_cat'’s own '.to_json()' presets), this function takes 'spec_toml' directly and parses it via ['AlgorithmSpec::from_toml'], the SAME entry point ‘sz_run_experiment_raw'’s 'ExperimentSpec::from_toml' already establishes the "hand a raw TOML document straight to the Rust core" convention for ('experiment.rs') – no R-side TOML library exists or is needed (r-sezgi has none in ‘DESCRIPTION'’s ‘Suggests'; unlike py-sezgi’s test, which parses TOML with the stdlib's own 'tomllib' into a dict before handing it to 'solve()', R has no such stdlib module, so parsing happens in Rust instead – 'AlgorithmSpec::from_toml'/'::from_json' are just two serializations of the identical schema, so this is not a private shortcut, only a different serialization entry point already used elsewhere in this same file's crate). Mirrors 'sz_solve_onemax' otherwise, EXCEPT 'run_id' is dropped (fixed to '0' internally) rather than taken as an explicit parameter – with 'spec_toml' this function already sits at 7 R-facing parameters; adding 'run_id' would push it to 8 and trip this workspace's 'clippy::too_many_arguments' gate (threshold 7, this file's ONE pre-existing exception is 'sz_preset_es_mu_plus_lambda_raw', not to be joined by a second). 'run_id = 0‘ matches how this scaffold is actually exercised (this task’s own TOML test, mirroring py-sezgi's 'test_gen_compound_mixed_space_toml_spec_ solves_end_to_end', calls 'solve(spec, problem, master_seed=42)' with no ‘run_id' override either – 'solve()'’s own Python signature defaults ‘run_id=0'). Test scaffolding only – NOT one of Task 5’s brief-pinned diagnostics, and (unlike onemax/int_quadratic/cat_match) has no verified target: this file's own convention never surfaces 'Problem::optimum()' in a result anyway (see ‘sz_solve_bbob'’s own 'best_f'/'evals'/'best_x' shape), so that caveat needs no separate plumbing here.

Description

Runs an algorithm spec on [‘MixedDiagnostic'] (this file’s own Float+Int+Categorical+Binary mixed-space scaffold problem, see its own doc) and returns the result – M3-8 Task 10. Added SOLELY so 'gen/compound' (Task 5) is reachable end to end through the NORMAL R solve path, proven with a mixed-space 'AlgorithmSpec' authored as TOML (this task's own test) – UNLIKE its three siblings above (which take ‘spec_json', pairing with 'sz_preset_ga_bin/ga_int/ga_cat'’s own '.to_json()' presets), this function takes 'spec_toml' directly and parses it via ['AlgorithmSpec::from_toml'], the SAME entry point ‘sz_run_experiment_raw'’s 'ExperimentSpec::from_toml' already establishes the "hand a raw TOML document straight to the Rust core" convention for ('experiment.rs') – no R-side TOML library exists or is needed (r-sezgi has none in ‘DESCRIPTION'’s ‘Suggests'; unlike py-sezgi’s test, which parses TOML with the stdlib's own 'tomllib' into a dict before handing it to 'solve()', R has no such stdlib module, so parsing happens in Rust instead – 'AlgorithmSpec::from_toml'/'::from_json' are just two serializations of the identical schema, so this is not a private shortcut, only a different serialization entry point already used elsewhere in this same file's crate). Mirrors 'sz_solve_onemax' otherwise, EXCEPT 'run_id' is dropped (fixed to '0' internally) rather than taken as an explicit parameter – with 'spec_toml' this function already sits at 7 R-facing parameters; adding 'run_id' would push it to 8 and trip this workspace's 'clippy::too_many_arguments' gate (threshold 7, this file's ONE pre-existing exception is 'sz_preset_es_mu_plus_lambda_raw', not to be joined by a second). 'run_id = 0‘ matches how this scaffold is actually exercised (this task’s own TOML test, mirroring py-sezgi's 'test_gen_compound_mixed_space_toml_spec_ solves_end_to_end', calls 'solve(spec, problem, master_seed=42)' with no ‘run_id' override either – 'solve()'’s own Python signature defaults ‘run_id=0'). Test scaffolding only – NOT one of Task 5’s brief-pinned diagnostics, and (unlike onemax/int_quadratic/cat_match) has no verified target: this file's own convention never surfaces 'Problem::optimum()' in a result anyway (see ‘sz_solve_bbob'’s own 'best_f'/'evals'/'best_x' shape), so that caveat needs no separate plumbing here.

Usage

sz_solve_mixed_diagnostic(
  spec_toml,
  n_float,
  n_int,
  k_cat,
  n_cat,
  n_bin,
  master_seed
)

Arguments

Value

A named list with 'best_f' (double), 'evals' (double), and 'best_x' – a MULTI-block genotype, surfaced as an unnamed list of 4 per-block vectors in 'SearchSpace::blocks()' order (Float numeric, Int integer, Categorical integer, Binary logical – see ‘genotype_to_r'’s doc).

# Errors A savvy error for any ['sezgi_core::spec'] parse error, or any ['sezgi_core::engine'] run error.

Runs an algorithm spec on [‘OneMax'] (Goldberg 1989’s classic Binary-block GA diagnostic; 'sezgi_problems::diagnostics::OneMax') and returns the result – M3-8 Task 10, mirroring 'sz_solve_tsp'/ 'sz_solve_cec2022' exactly ('Engine::from_spec' + 'engine.run'), except 'best_x' is now typed via ['genotype_to_r'] rather than assumed 'Float' (see that helper's own doc for the full type-mapping table). Pairs with 'sz_preset_ga_bin(...)'. Diagnostic only – not a benchmark, see ‘OneMax'’s own module doc.

Description

Runs an algorithm spec on [‘OneMax'] (Goldberg 1989’s classic Binary-block GA diagnostic; 'sezgi_problems::diagnostics::OneMax') and returns the result – M3-8 Task 10, mirroring 'sz_solve_tsp'/ 'sz_solve_cec2022' exactly ('Engine::from_spec' + 'engine.run'), except 'best_x' is now typed via ['genotype_to_r'] rather than assumed 'Float' (see that helper's own doc for the full type-mapping table). Pairs with 'sz_preset_ga_bin(...)'. Diagnostic only – not a benchmark, see ‘OneMax'’s own module doc.

Usage

sz_solve_onemax(spec_json, n_bits, master_seed, run_id)

Arguments

Value

A named list with 'best_f' (double), 'evals' (double), and ‘best_x' (a LOGICAL vector, one per bit – see ['genotype_to_r']’s doc).

# Errors A savvy error for any ['sezgi_core::spec'] parse error, or any ['sezgi_core::engine'] run error.

Runs an algorithm spec on a TSPLIB VENDORED instance ('"berlin52"', '"eil51"', '"st70"' – via ['Tsp::vendored']; UNLIKE 'sz_tsp_load()'/ 'sz_tsp_tour_length()' in 'problems.rs', raw TSPLIB text is not accepted here – mirrors py-sezgi's 'sezgi.problems.tsp(name)', which is likewise vendored-only) and returns the result. Same output shape as ‘sz_solve_bbob()' ('best_f'/'evals'/'best_x'), not py-sezgi’s own 'solve()' dict shape ('best_f'/'best_x'/'evals_used'/'iterations') – the established r-sezgi ‘sz_solve_*' convention governs here, not py-sezgi’s key names (see ‘problems.rs'’s module doc, "Index-convention decision", for the general 1-based-vs-0-based rule this function's 'best_x' also follows).

Description

Runs an algorithm spec on a TSPLIB VENDORED instance ('"berlin52"', '"eil51"', '"st70"' – via ['Tsp::vendored']; UNLIKE 'sz_tsp_load()'/ 'sz_tsp_tour_length()' in 'problems.rs', raw TSPLIB text is not accepted here – mirrors py-sezgi's 'sezgi.problems.tsp(name)', which is likewise vendored-only) and returns the result. Same output shape as ‘sz_solve_bbob()' ('best_f'/'evals'/'best_x'), not py-sezgi’s own 'solve()' dict shape ('best_f'/'best_x'/'evals_used'/'iterations') – the established r-sezgi ‘sz_solve_*' convention governs here, not py-sezgi’s key names (see ‘problems.rs'’s module doc, "Index-convention decision", for the general 1-based-vs-0-based rule this function's 'best_x' also follows).

Usage

sz_solve_tsp(spec_json, name, master_seed, run_id)

Arguments

Value

A named list with 'best_f' (double), 'evals' (double), and 'best_x' (double vector – a 1-based permutation of '1:n_cities', the best EVALUATED tour, paired with 'best_f').

# Errors A savvy error if 'name' is not one of the three vendored instances, for any ['sezgi_core::spec'] parse error, or any ['sezgi_core::engine'] run error.

Composes one or more blocks into a search space, in the given order.

Description

Mirrors 'sezgi.Space' ('py-sezgi/python/sezgi/spaces.py') exactly: at least one block is required, and every argument must inherit '"sz_block"' (i.e. be built by [sz_float()], [sz_int()], [sz_categorical()], [sz_binary()], or [sz_permutation()]) – these are the only two validation rules 'Space.__init__' itself performs.

Usage

sz_space(...)

Arguments

Value

An object of class '"sz_space"' with element 'blocks' (a 'list' of the given blocks, in order).

Examples

sz_space(sz_float(-5, 5, 3), sz_int(0L, 10L, 2))

Bayesian signed-rank test with a region of practical equivalence (ROPE).

Description

Mirrors py-sezgi's 'stats_bayesian_signed_rank()'. Dirichlet-weighted Monte Carlo scheme (Benavoli et al. 2017 simplification); see 'crates/stats/src/bayesian.rs' for the full algorithm and determinism guarantees. The same '(a, b, rope, samples, seed)' always produces a bit-identical result – R calls the exact same seeded Rust core as Python and Rust.

Usage

sz_stats_bayesian_signed_rank(a, b, rope = 0, samples = 20000, seed = 1)

Arguments

Value

A named list with 'p_left', 'p_rope', 'p_right'.

Cliff's delta effect size for two independent (unpaired) samples.

Description

Cliff's delta effect size for two independent (unpaired) samples.

Usage

sz_stats_cliffs_delta(a, b)

Arguments

Value

A numeric scalar in '[-1, 1]'.

Qualitative magnitude label for a Cliff's delta value (Romano et al. 2006 thresholds).

Description

Qualitative magnitude label for a Cliff's delta value (Romano et al. 2006 thresholds).

Usage

sz_stats_cliffs_magnitude(delta)

Arguments

Value

A character scalar: one of '"negligible"', '"small"', '"medium"', '"large"'.

Runs the Friedman test on a results matrix.

Description

Runs the Friedman test on a results matrix.

Usage

sz_stats_friedman(m)

Arguments

Value

A named list with 'statistic', 'p_value', 'mean_ranks' (mirrors py-sezgi's 'stats_friedman()' dict keys exactly).

Comprehensive statistical analysis package for algorithm comparison.

Description

Mirrors py-sezgi's 'stats_paper_package()': Friedman test, Nemenyi critical difference, all-pairs Wilcoxon signed-rank with Holm adjustment, Cliff's delta, Bayesian signed-rank, Plackett-Luce ranking, and LaTeX summary/pairwise-comparison tables. See 'crates/stats/src/report.rs'.

Usage

sz_stats_paper_package(
  algo_names,
  problem_names,
  m,
  rope = 0,
  samples = 20000,
  seed = 1
)

Arguments

Value

A named list with 'friedman', 'nemenyi_cd', 'pairwise_wilcoxon_holm', 'cliffs', 'bayes', 'plackett_luce', 'latex_summary', 'latex_tests' – see 'sz_stats_paper_package_raw()' for the full field-by-field shape.

Plackett-Luce maximum-likelihood ranking (Hunter 2004 MM algorithm; see 'sezgi_stats::plackett_luce').

Description

Plackett-Luce maximum-likelihood ranking (Hunter 2004 MM algorithm; see 'sezgi_stats::plackett_luce').

Usage

sz_stats_plackett_luce(rankings)

Arguments

Value

A named list with 'worths', 'p_best', 'iterations' (mirrors py-sezgi's 'stats_plackett_luce()' dict keys exactly).

Wilcoxon signed-rank test for two paired samples (Pratt zero-handling and tie correction; see 'sezgi_stats::wilcoxon_signed_rank').

Description

Wilcoxon signed-rank test for two paired samples (Pratt zero-handling and tie correction; see 'sezgi_stats::wilcoxon_signed_rank').

Usage

sz_stats_wilcoxon(a, b)

Arguments

Value

A named list with 'w_statistic', 'z', 'p_value', 'n_effective', ‘method' ('"exact"' or '"normal_approx"'; mirrors py-sezgi’s 'stats_wilcoxon()' dict keys exactly). '"exact"' is used when 'n_effective <= 25' and there are no zero differences or tied '|d|' ranks; see ‘sezgi_stats::wilcoxon_signed_rank'’s doc comment for the full eligibility rule (the exact p-value formula is semver-pinned).

Examples

sz_stats_wilcoxon(c(1, 2, 3, 4, 5), c(2, 1, 4, 3, 6))

Loads a TSPLIB 'EUC_2D' instance, either a vendored instance name ('"berlin52"', '"eil51"', '"st70"') or raw TSPLIB file text (see ['load_tsp']).

Description

Loads a TSPLIB 'EUC_2D' instance, either a vendored instance name ('"berlin52"', '"eil51"', '"st70"') or raw TSPLIB file text (see ['load_tsp']).

Usage

sz_tsp_load(name_or_text)

Arguments

Value

A named list with 'name' (character scalar), 'n_cities' (double), ‘coords' (an 'n_cities' x 2 numeric matrix, row 'i' is city 'i'’s ‘(x, y)' coordinate pair – see this module’s own doc, "Index-convention decision"), 'known_optimum' (double, or 'NULL' for an instance parsed from raw text rather than a vendored name).

# Errors A savvy error for any ['TspError'] (unknown vendored name that also fails to parse as TSPLIB text, malformed TSPLIB text, unsupported 'EDGE_WEIGHT_TYPE', ...).

Examples

tsp <- sz_tsp_load("berlin52")
tsp$n_cities

Closed-tour length of a 1-based 'tour' (a permutation of '1:n_cities' – see this module's own doc, "Index-convention decision") on the instance named/parsed by 'name_or_text' (see ['load_tsp']), via [‘Tsp::evaluate_batch']’s ‘nint'-rounded 'EUC_2D' sum ('tsp.rs'’s module doc).

Description

UNLIKE 'Tsp::evaluate_batch' itself (which returns 'f64::INFINITY' for a malformed genotype, since genotype validity is normally the engine's ‘SearchSpace::validate' job, not 'Tsp'’s own) – this binding validates 'tour' itself (exact length, every entry in '1..=n_cities', no repeats) and raises a precise error instead, since a 'tour' coming directly from R has no engine-side validation gate in front of it.

Usage

sz_tsp_tour_length(name_or_text, tour)

Arguments

Value

A numeric scalar.

# Errors A savvy error for any ['TspError'] resolving 'name_or_text', or if 'tour' is not a permutation of '1:n_cities' (wrong length, an out-of-range entry, or a repeated entry).