FullDatasetEvaluationStrategy
Fit the model once on every row and score it on those same rows.
For a task with no target, such as clustering, nothing can be held out:
there is no y_true to compare a prediction against, and the metrics are
internal ones computed from the features and the labels the model assigned
them. So the dataset is not carved at all, and the run records a single
FULL score per metric.
A run carved this way is prepared by PrepareWithoutTargetUnit, fitted
by FitWithoutTargetUnit and scored by ScoreClustersUnit. It takes no
splitter and no hyperparameter search, since a search needs a held-out
partition to be measured on.
Only offered for tasks without a target. The session API enforces the pairing, which is what lets a run of this kind assume there is no target.
Methods
get_metadata(cls) -> dict
BaseEvaluationStrategyDescribe the strategy for the frontend.
Returns
- dict
- Mapping with
kind, which says whether this strategy splits the dataset once or into folds, andscored_splits, the partitions it writes metrics for.