ICLEstimator.score()

Submits an ICL scoring job and returns a JobMonitor. Runs the same pipeline as predict() but on the validation split, then computes every metric available for the task type against known labels — not just the one named in task.evaluation_metric. Since there is no training loop to justify tracking a single metric, and the rest are nearly free to compute once predictions exist, score() returns the full set.

Parameters

Name	Type	Description	Optional
dataset	RelationalDataset | TabularDataset	Dataset to score.	No
val_batch_size	int	Number of rows per inference chunk. Defaults to 256 (see predict()’s test_batch_size for why).	Yes
val_table	str	File/table path for the validation split. Optional — falls back to the task’s own validation_table when not provided.	Yes

Returns

A JobMonitor. Once the job completes, retrieve the computed metrics via score_job.get_status()["result"], which contains:

• metric_name — the task’s configured evaluation_metric.
• metrics — a dict of every metric computed for the task type (e.g. for binary_classification: roc_auc, plus the other classification metrics valid for that task type), keyed by metric name. Look up metrics[metric_name] for the value of the configured metric.

Requirements

Same as predict():

• task.label_column must be set.
• task.context_table must be set.
• The task’s validation_table (or the val_table argument) must exist and contain the required label/entity/time columns.
• If ICLConfig.sampling_strategy is "most_recent" or "mixed", the task must have a time_column set.

Examples

Relational dataset

from relationalai_predictive import ICLEstimator, ICLConfig

model = ICLEstimator(connector=connector, config=ICLConfig())

score_job = model.score(relational_dataset)
result = score_job.get_status()["result"]

print(result["metric_name"], result["metrics"][result["metric_name"]])
print(result["metrics"])  # every metric for this task_type

Flat / forecasting dataset

from relationalai_predictive import ICLEstimator, TabularTask, TabularDataset, TaskType

task = TabularTask(
    name="sales_forecast",
    task_type=TaskType.FORECASTING,
    label_column="sales",
    context_table="DATABASE.SCHEMA.SALES_CONTEXT",
    validation_table="DATABASE.SCHEMA.SALES_VALIDATION",
    time_column="date",
    prediction_length=12,
    max_context_length=64,
)
dataset = TabularDataset(connector=connector, name="sales", task=task)

model = ICLEstimator(connector=connector)
score_job = model.score(dataset)

Common Validation Errors

See predict() — score() shares the same label_column/context_table/sampling_strategy validation, but checks the validation split’s columns instead of the test split’s.

• “Task ’…’ has no ‘validation’ split” / “TabularTask ’…’ has no ‘validation’ split” — set validation_table= on the task, or pass val_table= to score().