OPTIMAL PROGNOSIS (PHP-ONLY)¶
Trains all PHP-side candidate surrogates (polynomial orders 1–3 and Kriging) on a numeric dataset, scores each with K-fold cross-validation, and returns a leaderboard ranked by Coefficient of Prognosis (out-of-fold R²). Use this worker for fast, local surrogate selection on small-to-medium datasets without requiring cloud training resources.
When to use¶
Tagged: cop, cross-validation, kriging, local, lucy_ml, mop, php, polynomial.
Inputs¶
| Label | ID | Type | Default | Required | Description |
|---|---|---|---|---|---|
| Dataset | dataset | dataset | — | ✓ | Tabular training dataset (rows of associative arrays) from which features and the target column are drawn; must contain at least 4 rows. |
| Input Features | independents | select | — | ✓ | One or more numeric column names from the dataset to use as input features (predictors) for all candidate surrogates. |
| Target Column | clabel | select | — | ✓ | Name of the single numeric column in the dataset to be used as the regression target (response variable). |
| Cross-Validation Folds | kfolds | text | 5 | Number of folds for K-fold cross-validation (integer); defaults to 5 — increase for larger datasets to obtain tighter Coefficient of Prognosis estimates. | |
| Shuffle Seed | seed | text | — | Optional integer seed for the fold-shuffle RNG; leave blank for a non-deterministic run or set to any integer for fully reproducible CV splits. | |
| Candidate Surrogates | surrogates | select | (complex) | Subset of candidate surrogate types to compete (polynomial_o1, polynomial_o2, polynomial_o3, kriging_gaussian); defaults to all four — narrow the list to skip slow models on very small datasets. |
Outputs¶
| Label | ID | Type | Description |
|---|---|---|---|
| Leaderboard | leaderboard | dataset | Dataset (mop_leaderboard) with one row per candidate surrogate, containing CoP, RMSE, MAE, in-sample R², elapsed time, and status, sorted by CoP descending with failed runs at the bottom. |
| Winning Surrogate | winner | text | String identifier of the surrogate with the highest Coefficient of Prognosis (e.g. ‘kriging_gaussian’ or ‘polynomial_o2’); empty string if all candidates failed. |
| Winner Coefficient of Prognosis | winner_cop | scalar | Scalar Coefficient of Prognosis (out-of-fold R², range 0–1) for the winning surrogate; null if no surrogate succeeded. |
| Winner Predictions (out-of-fold) | winner_predictions | dataset | Dataset (mop_predictions) of out-of-fold predictions for the winning surrogate, with columns row_index, actual, predicted, and residual — useful for residual-analysis plots. |
| Prognosis Summary | mop_summary | json | JSON blob (mop_summary) containing the full run configuration (inputs, target, n_rows, kfolds, seed, candidate list) and per-surrogate fold-level scores, suitable for audit trails or downstream reporting. |
Disciplines¶
- ai_ml.model_selection
- ai_ml.prognosis
- ai_ml.supervised.regression
- ai_ml.surrogate
Auto-generated from platform schema. Worker id: ml_learn_mop. Schema hash: 06a97cbb24fb. Hand-curated docs in workerexamples/ override this page when present.