Skip to content

Scenarios

Scenario files live in examples/scenarios/. They are versioned ExperimentSpec JSON payloads accepted by the CLI and dashboard API.

Run one from the command line:

uv run pspso run start examples/scenarios/breast_cancer_svm_random.json --wait

Or reproduce it in the dashboard by selecting the same dataset, task, estimator, strategy, fixed params, and tunable params.

Included Scenarios

File Dataset Task Estimator Strategy
breast_cancer_svm_random.json Breast cancer Binary classification SVM Random
breast_cancer_random_forest_grid.json Breast cancer Binary classification RandomForest Grid
diabetes_svm_regression_random.json Diabetes Regression SVM Random
diabetes_random_forest_pso.json Diabetes Regression RandomForest PSO
csv_binary_classification_template.json CSV Binary classification SVM Random
csv_regression_template.json CSV Regression RandomForest Random
wine_random_forest_multiclass.json Wine Multiclass classification RandomForest Random

Dashboard Reproduction

For a checked-in scenario:

  1. Open the JSON file.
  2. Match the dataset fields in the dashboard.
  3. Select the same task, metric, estimator, and strategy.
  4. Use the same fixed params and tunable params.
  5. Validate.
  6. Start the run only after validation succeeds.

Multiclass and PyTorch

The Wine scenario demonstrates macro F1 and a reproducible multiclass run:

uv run pspso run start examples/scenarios/wine_random_forest_multiclass.json --wait

The optional PyTorch tabular MLP emits epoch events for the live timeline:

uv sync --extra torch

The frontend does not execute arbitrary code. Select a registered model recipe or use the documented Python extension contract for a custom estimator.