SALib: Sensitivity Analysis Library

1.5.2 · active · verified Tue Apr 14

SALib is an open-source Python library providing implementations of various global sensitivity analysis methods, including Sobol', Morris, FAST, DGSM, PAWN, HDMR, Moment Independent, and fractional factorial. It is currently at version 1.5.2 and maintains an active development status with somewhat regular releases.

Warnings

breaking SALib version 1.2 and onwards officially dropped support for Python 2. Users must use Python 3 or later versions.
Fix: Ensure your Python environment is running Python 3.x (>=3.9 recommended for 1.5.2).
breaking Starting with SALib 1.0.0, the Python API for sampling and analysis methods expects NumPy matrices directly as input for model runs and results, rather than file paths. The command-line interface remains unchanged.
Fix: Update your code to pass NumPy arrays (e.g., `param_values`, `Y`) directly to sampling and analysis functions like `saltelli.sample()` and `sobol.analyze()`.
breaking SALib 1.5.2 introduced an updated NumPy dependency, requiring NumPy version 2.0 or higher.
Fix: Upgrade NumPy in your environment to version 2.0 or newer using `pip install --upgrade numpy`.
gotcha Mismatching sampling and analysis methods can lead to incorrect sensitivity indices. For example, Sobol' analysis requires samples generated by a Sobol' sampler (like Saltelli's).
Fix: Always use the appropriate sampling method for the chosen analysis method, as specified in the SALib documentation. E.g., `SALib.sample.saltelli` for `SALib.analyze.sobol`.
gotcha Sobol' index estimates can be biased if model outputs are not centered. SALib normalizes outputs by default to mitigate this. If using non-normalized outputs, larger sample sizes may be required for convergence.
Fix: Be aware of the output normalization behavior if comparing results with other tools, and consider increasing sample size for non-centered outputs if normalization is bypassed or if convergence issues are observed.

Install

pip install SALib PyPI (recommended)
conda install conda-forge::salib Conda

Imports

sample_method
```
from SALib.sample import saltelli
```
Sampling methods are imported from the `SALib.sample` submodule.
analyze_method
```
from SALib.analyze import sobol
```
Analysis methods are imported from the `SALib.analyze` submodule.
ProblemSpec
```
from SALib import ProblemSpec
```
For the object-oriented interface, ProblemSpec is imported directly from the top-level `SALib` package.

Quickstart

This quickstart demonstrates a Sobol' sensitivity analysis using the Ishigami test function. It covers defining problem parameters, generating samples using the Saltelli sampler, evaluating a model, and analyzing the results to compute sensitivity indices.

import numpy as np
from SALib.sample import saltelli
from SALib.analyze import sobol
from SALib.test_functions import Ishigami

# 1. Define the model inputs
problem = {
    'num_vars': 3,
    'names': ['x1', 'x2', 'x3'],
    'bounds': [[-np.pi, np.pi]] * 3
}

# 2. Generate samples
param_values = saltelli.sample(problem, 1024)

# 3. Run model (example: Ishigami function)
Y = Ishigami.evaluate(param_values)

# 4. Perform analysis
Si = sobol.analyze(problem, Y, print_to_console=True)

print(f"First-order sensitivity indices: {Si['S1']}")
print(f"Total-order sensitivity indices: {Si['ST']}")

view raw JSON →