pasys optimize¶

Optimize a design over the config's design space, either scalarized (scipy) or multi-objective (NSGA-II).

Synopsis¶

pasys optimize <config> --objective <metric> [options]

Description¶

The optimize command searches the design space defined in the config's doe.variables section. With the scipy methods it minimizes or maximizes a single scalarized objective; with --method nsga2 it returns the full Pareto front for up to two objectives (requires the [mdao] extra). Requirements in the config act as constraints: normalized penalties for the scipy methods, constraint domination for NSGA-II.

Arguments¶

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

Options¶

Option	Default	Description
`--objective`	(required)	Metric key to optimize
`--sense`	`maximize`	`maximize` or `minimize`
`--method`	`de`	`de`, `da`, `minimize`, or `nsga2`
`--objective2`		Second objective for nsga2, as `metric:sense`
`--max-iter`	`200`	Maximum iterations (scipy methods)
`--generations`	`100`	NSGA-II generations
`--population`	`50`	NSGA-II population size
`--seed`	`42`	Random seed
`--output`, `-o`		Save result (JSON; Parquet for nsga2)

Methods¶

de — differential evolution: global, handles integer variables natively; the default and the recommended scipy method.
da — dual annealing: global, continuous relaxation of integers.
minimize — L-BFGS-B: local; only useful for all-continuous spaces.
nsga2 — multi-objective NSGA-II via pymoo: returns the nondominated set with full metrics per point instead of a single design. Install with pip install "phased-array-systems[mdao]".

Examples¶

Scalarized run:

pasys optimize config.yaml --objective eirp_dbw --sense maximize --method de

Pareto front for EIRP vs cost:

pasys optimize config.yaml --objective eirp_dbw --method nsga2 \
    --objective2 cost_usd:minimize --generations 100 --population 50 \
    -o pareto.parquet

The Parquet output has the same schema as DOE results, so pasys report and pasys pareto work on it directly.