Results¶

Pareto-optimal points and optimization metadata.

Summary analysis computed automatically after optimization.

Results keyed by scenario name.

Cross-scenario comparison metrics.

Pareto-optimal points satisfying all hard constraints.

Points that violated at least one hard constraint.

Whether any feasible solutions were found.

Index into feasible_points of the recommended compromise solution (multi-objective only).

Hypervolume indicator of the Pareto front (multi-objective only).

The problem definition used for optimization.

Number of generations executed.

Total number of candidate evaluations.

Decision and context variable values.

Predicted objective values by name.

Constraint evaluations with violation info.

Whether all hard constraints are satisfied.

Distance to the training data manifold.

Whether this point is outside the training domain.

Name of the constraint.

Constraint violation magnitude (0.0 if satisfied).

Predicted value from the surrogate.

Hard or soft constraint.

Lower conformal prediction bound (data constraints only).

Upper conformal prediction bound (data constraints only).

High-level summary including solution overview, baseline comparison, constraint status, surrogate quality, and extrapolation warnings.

Overview of feasible/infeasible solutions and best objectives.

Improvement vs. historical best (None if no feasible solutions).

Status of each constraint at the recommended point.

Quality metrics per trained surrogate.

Feasible points flagged as extrapolating.

Detected violations of declared monotonic relations.

Number of feasible Pareto-optimal points.

Number of infeasible points.

Best value found per objective across feasible points.

Objective values at the compromise point (multi-objective only).

Hypervolume indicator of the Pareto front (multi-objective only).

Predicted objectives at the recommended point.

Best historical value per objective from the training data.

Improvement over historical best (positive = better).

Raw constraint evaluation with violation and prediction.

Classification as satisfied, active, or violated.

Distance to the constraint boundary (positive = satisfied).

Target column name.

Best cross-validation RMSE from the Optuna study.

Conformal prediction interval coverage level.

Number of models in the ensemble.

Quality warning if CV RMSE significantly exceeds conformity scores.

Index into the feasible points list.

k-NN distance to the training data manifold.

Variable involved in the violation.

Objective or constraint target.

Expected monotonic direction.

Fraction of grid intervals where monotonicity is violated.

Maximum absolute reversal magnitude.

Target column name.

Mean |SHAP| per feature (all variables).

Mean |SHAP| per feature (decision variables only).

Target column name.

Feature names in column order.

SHAP value matrix, shape (n_samples, n_features).

Expected model output (ensemble-weighted).

Feature value matrix, shape (n_samples, n_features).

Target column name.

Feature names in column order.

SHAP values for this point, shape (n_features,).

Expected model output (ensemble-weighted).

Feature values at this point.

Model prediction (base_value + sum of SHAP values).

Decision variable being varied.

Target column being predicted.

Grid points along the variable range, shape (n_grid,).

Ensemble-weighted PDP values, shape (n_grid,).

Ensemble-weighted ICE curves, shape (n_samples, n_grid).

All pairwise combinations of objectives.

Marginal rates of substitution per objective pair, each an array of rates between adjacent Pareto points.

Objective values on the Pareto front, shape (n_points, n_objectives), sorted by first objective.

The hypothetical variable values that were evaluated.

Predictions per objective.

Predictions per data constraint.

Distance to the training data manifold.

Whether this point is outside the training domain.

Point prediction from the surrogate ensemble.

Lower conformal prediction interval bound.

Upper conformal prediction interval bound.

Prediction at the recommended (compromise) point for comparison.

Mean of the historical data for this column.

Names of the compared scenarios.

Robustness assessment per decision variable.

Name of the decision variable.

Recommended value per scenario.

Whether the variable is stable across scenarios (spread < 5% of bounds range).

Absolute range of values across scenarios (0.0 for categorical agreement).