Skip to content

Campaign workflow

A campaign is a Python recipe with four moving parts: case, evaluator, budget and archive.

1. Define the case

from aster.core import CaseSpec, Constraint, FamilySpec, Objective, ParamSpec, ParamSpace

case = CaseSpec(
    name="ring_demo",
    family=FamilySpec("ring", description="radial ring on a ground plane"),
    space=ParamSpace([
        ParamSpec("ring_radius", "mm", 20.0, 60.0, 40.0),
        ParamSpec("ring_height", "mm", 1.0, 10.0, 4.0),
        ParamSpec("slot_depth", "mm", 0.0, 6.0, 2.0),
    ]),
    objectives=[
        Objective("gain", "peak_gain", "maximize"),
        Objective("bw", "s11_bandwidth", "maximize"),
    ],
    constraints=[Constraint("bw", ">=", 5.0)],
    project="<path to reference .aedt>",
    setup="Setup1",
)

The built-in cases under aster/cases/ are references for real designs; their project paths are left empty on purpose and must be configured for your own checkout.

2. Pick an evaluator

  • Dry evaluator — a fitted measured surface or an analytic response. It is fast and is labelled PREDICTED; it is useful for wiring the loop and for smoke tests.
  • Real evaluator — submits the candidate to the guard runner, solves, exports S11 and far field, and writes the metrics from the artifacts. It needs AEDT and the guard checkout.

3. Run and iterate

candidates = latin_hypercube(case.space, 16, seed=1)
evaluations = [evaluator.evaluate(candidate) for candidate in candidates]
archive = pareto_archive(evaluations, case.objectives)

For longer studies, use the campaign driver:

python -m aster.worker --case ring_demo --campaign <campaign-dir> --generations 8

The worker persists every run in campaign.sqlite with its parameters, metrics, status and artifact paths.

4. Re-score without re-solving

Recorded runs carry their raw artifacts. A changed acceptance rule (for example a different bandwidth definition) is applied to the archive with band_acceptance/pattern_acceptance and the new verdicts are written next to the old ones — the solver is not called again.

Acceptance rules

Use explicit verdict functions rather than ad-hoc thresholds. band_acceptance.assess_touchstone_band and pattern_acceptance.assess_ring_pattern return a structured result with the measured values, the rule and a pass/fail decision; acceptance.py combines them per run.