Warning, /AID2E-framework/examples/dtlz2/README.md is written in an unsupported language. File is not indexed.
0001 # DTLZ2 Examples
0002
0003 [dtlz2_ax_joblib.yml](configurations/dtlz2_ax_joblib.yml) configures Ax optimization with 10 design variables, 2 minimized objectives, and JobLib execution.
0004
0005 From the repository root:
0006
0007 ```bash
0008 aid2e describe examples/dtlz2/configurations/dtlz2_ax_joblib.yml
0009 aid2e optimize examples/dtlz2/configurations/dtlz2_ax_joblib.yml --validate-only
0010 aid2e optimize examples/dtlz2/configurations/dtlz2_ax_joblib.yml
0011 ```
0012
0013 See [configurations](configurations/) for optimization settings, [python_api](python_api/) for direct API examples, and [schedulers](schedulers/) for Slurm and PanDA examples. The [configuration guide](../../docs/user-guide/configuration.md) explains the supported schema.
0014
0015 [dtlz2_design.params](dtlz2_design.params) provides the shared unconstrained design with ten variables. [dtlz2_evaluator.py](dtlz2_evaluator.py) supports Python callables, direct `--x1` through `--x10` arguments, and `--design` JSON input. PanDA stage and dataset examples use a separate evaluator.
0016
0017 ## Configuration examples
0018
0019 | Configuration | Feature |
0020 | --- | --- |
0021 | [dtlz2_ax_joblib.yml](configurations/dtlz2_ax_joblib.yml) | Bayesian optimization with local trial batches. |
0022 | [dtlz2_ax_joblib.json](configurations/dtlz2_ax_joblib.json) | The same configuration in JSON. |
0023 | [dtlz2_ax_slurm.yml](configurations/dtlz2_ax_slurm.yml) | Slurm submission and command-based evaluation. |
0024 | [dtlz2_ax_panda.yml](configurations/dtlz2_ax_panda.yml) | PanDA/iDDS callable submission. |
0025 | [dtlz2_ax_constraints.yml](configurations/dtlz2_ax_constraints.yml) | Inline design parameters and linear Ax constraints. |
0026 | [dtlz2_objective_collection.yml](configurations/dtlz2_objective_collection.yml) | Extract both objectives from one script execution. |
0027 | [dtlz2_objective_plans.yml](configurations/dtlz2_objective_plans.yml) | Separate objective plans, script/inline steps, and dependencies. |
0028 | [dtlz2_pymoo_joblib.yml](configurations/dtlz2_pymoo_joblib.yml) | NSGA-II population and offspring evaluation. |
0029 | [dtlz2_scheduler_cascade.yml](configurations/dtlz2_scheduler_cascade.yml) | Global, workflow, branch, and stage scheduler precedence. |
0030 | [dtlz2_sobol_joblib.yml](configurations/dtlz2_sobol_joblib.yml) | Sobol sampling without Bayesian generation. |
0031
0032 The Ax, PyMOO, and Sobol JobLib baselines share the same unconstrained design and 100-evaluation budget. Ax uses 20 initial Sobol samples followed by 80 Bayesian candidates, in batches of four. PyMOO uses a population of 20 followed by 20 generations of four offspring; its `n_iterations` counts generations. These are example settings, not a convergence or performance guarantee. Slurm and PanDA require site-specific setup before execution.
0033
0034 The Python optimizer scripts and notebook reuse the Ax/PyMOO JobLib configs, using only their problem and optimizer sections. Slurm design-JSON and PanDA multistage variants remain for separate review.
