pasys sensitivity¶

Rank design-variable influence on output metrics, one-at-a-time or with Sobol global indices.

Synopsis¶

pasys sensitivity <config> [options]

Description¶

The sensitivity command sweeps the variables in the config's doe.variables section. The default one-at-a-time (OAT) mode sweeps each variable across its range while holding the others at the architecture baseline. Sobol mode (--sens-method sobol, requires the [mdao] extra) computes variance-based S1 (first-order) and ST (total, including interactions) indices from a Saltelli sample.

Arguments¶

Argument	Required	Description
`config`	Yes	Path to configuration file (YAML or JSON)

Options¶

Option	Short	Default	Description
`--sens-method`		`oat`	`oat` or `sobol`
`--steps`	`-n`	`5`	Steps per parameter (OAT)
`--samples`		`256`	Sobol base sample count (total runs = samples × (2D + 2))
`--metric`		`g_peak_db`	Metric to analyze
`--plot`			Generate tornado plots (OAT)
`--output`	`-o`		Save results (CSV/Parquet)

Choosing a method¶

OAT is cheap and shows local trends, but misses interactions between variables. Sobol attributes shares of the output variance to each input (and, via ST−S1, to its interactions), at the cost of many more evaluations.

Sobol needs a near-rectangular feasible domain: variables constrained by the sub-array rules (like array.nx) should stay fixed in the config's architecture section rather than being swept.

Examples¶

# One-at-a-time tornado analysis
pasys sensitivity config.yaml -n 7 --metric link_margin_db --plot

# Sobol global indices
pasys sensitivity config.yaml --sens-method sobol --samples 256 \
    --metric eirp_dbw -o sobol.parquet