Crate Map

salib is a workspace of nine crates. The salib facade re-exports the most common types and functions; depend on individual crates for finer control over compile times and dependencies.

Dependency graph

salib  (facade)
├── salib-core           types, distributions, RngState, deterministic reductions
├── salib-samplers       sampling designs (LHS, Sobol', Saltelli, Morris, FAST)
│   └── salib-core
├── salib-estimators     all sensitivity estimators
│   ├── salib-core
│   └── salib-samplers
├── salib-surrogate      PCE, HDMR, active subspaces  [optional: "surrogate"]
│   └── salib-core
├── salib-shapley        Shapley effects              [optional: "shapley"]
│   ├── salib-core
│   ├── salib-samplers
│   └── salib-estimators
├── salib-validation     test functions (Ishigami, Sobol' G, Morris)  [optional: "validation"]
│   └── salib-core
├── salib-tck            Gherkin test harness          [dev only]
└── salib-cli            CLI binary                    [separate install]
    ├── salib-core
    ├── salib-samplers
    └── salib-estimators

Crates

salib-core

Types that everything else depends on.

Type	Role
`Problem`	Factor names, distributions, dimension
`ProblemBuilder`	Fluent builder for `Problem`
`Factor`	Name + distribution for one input
`Distribution`	`Uniform`, `Normal`, `LogNormal`, `Triangular`, `Beta`
`RngState`	Deterministic ChaCha20 RNG with `split()` for parallelism
`tree_sum`, `tree_dot`, `tree_var`	Binary-tree accumulation for bit-deterministic parallel sums

salib-samplers

Sampling designs that produce the input matrices for estimators.

Design	Function / Type
Latin Hypercube	`LhsSampler`
Sobol’ QMC	`SobolSampler`
Saltelli cross-matrix	`build_saltelli_matrix`
Morris trajectories	`build_morris_trajectories`
FAST/eFAST search curves	`build_fast_design`
Iman–Conover correlation	`iman_conover_transform`

salib-estimators

All sensitivity estimators. Each takes a sample matrix (or design) and a model closure, returns a result struct with Display and optional serde.

Family	Function
Sobol’ (Saltelli 2010)	`estimate_saltelli2010`
Sobol’ (Jansen)	`estimate_jansen`
Sobol’ (Janon)	`estimate_janon`
Sobol’ (Owen)	`estimate_owen`
Given-data Sobol’	`estimate_given_data_sobol`
Morris	`estimate_morris_effects`
Grouped Morris	`estimate_grouped_morris_effects`
FAST / eFAST	`estimate_fast`
RBD-FAST	`estimate_rbd_fast`
Borgonovo δ	`estimate_borgonovo_delta`
PAWN	`estimate_pawn`
QOSA	`estimate_qosa`
DGSM	`estimate_dgsm`
SRC / SRRC / PCC / PRCC	`estimate_regression_indices`
HDMR	`estimate_hdmr`
ANOVA	`estimate_anova_two_way`
G-Theory	`estimate_g_theory_pir`
Discrepancy	`compute_discrepancy`
Fractional factorial	`estimate_fractional_factorial`

salib-surrogate

Surrogate models. Optional — enable with features = ["surrogate"] on the facade.

Component	Function
Full PCE (OLS)	`fit_full_pce`
Sparse PCE (LARS/OMP)	`fit_sparse_pce`
Active subspaces	`compute_active_subspace`

salib-shapley

Shapley effects. Optional — enable with features = ["shapley"].

Component	Function
Shapley effects	`estimate_shapley`

salib-validation

Analytic test functions with closed-form sensitivity indices. Optional — enable with features = ["validation"].

Function	Factors	Closed-form
Ishigami	3	S_i, S_{Ti}
Sobol’ G	d	S_i, S_{Ti}
Morris	d	\mu^*, \sigma classification
Linear	d	exact SRC

salib-tck

Internal test-contract-kit. Gherkin-based integration tests. Not published for external use.

salib-cli

Command-line interface. Install with cargo install salib-cli.

salib sample   # generate samples to CSV
salib run      # evaluate a model on samples
salib analyze  # compute sensitivity indices

Feature flags

All flags on the salib facade crate:

Flag	Default	Effect
`samplers`	yes	Sampling designs
`estimators`	yes	All sensitivity estimators
`parallel`	yes	Rayon-based parallel reductions
`surrogate`	no	PCE, HDMR, active subspaces
`shapley`	no	Shapley effects
`validation`	no	Analytic test functions
`serde`	no	`Serialize`/`Deserialize` on result types
`arrow`	no	Arrow `RecordBatch` conversions
`polars`	no	Polars `DataFrame` conversions
`full`	no	Everything except `serde`, `arrow`, `polars`

# Kitchen sink
salib = { version = "0.1", features = ["full", "serde"] }

# Minimal: just Sobol' indices, no parallelism
salib = { version = "0.1", default-features = false, features = ["samplers", "estimators"] }