gds-psuu v0.2.0

What's New

Composable Metric/Aggregation/KPI

Metric (per-run scalar) + Aggregation (cross-run reducer) = KPI. Built-in factories:

• Metrics: final_value, trajectory_mean, max_value, min_value
• Aggregations: mean_agg, std_agg, percentile_agg, probability_above, probability_below

Per-run distributions tracked in EvaluationResult.distributions.

Parameter Constraints

LinearConstraint and FunctionalConstraint define feasible regions. Grid search filters; random search uses rejection sampling.

Composable Objectives

SingleKPI and WeightedSum for multi-KPI optimization via SweepResults.best_by_objective().

Sensitivity Analysis

OATAnalyzer (one-at-a-time) and MorrisAnalyzer (elementary effects) for parameter importance screening.

Optuna Migration

Replaced unmaintained scikit-optimize with optuna>=4.0 for Bayesian optimization.

Documentation

Full docs: overview, getting started, concepts, parameter spaces, optimizers, and API reference.

Stats

• 127 tests, 91% coverage
• Zero breaking changes (legacy KPI(fn=...) still works)

Install

uv add gds-psuu
# or: pip install gds-psuu