Scores

Scores(self, values, tool_names, metric_ids, dataset_names, layout, source_path=None, blinding_sha256=None)

A benchmark score table or tensor with its tool and metric labels.

Holds either a wide tool by metric matrix (layout == "wide", values is 2D and dataset_names is None) or a long tool by dataset by metric tensor (layout == "long", values is 3D). The metric ids have all been checked against the registry at load time.

Attributes

Name	Type	Description
values	np.ndarray	Float array. Shape (n_tools, n_metrics) when wide, or (n_tools, n_datasets, n_metrics) when long. Missing cells are numpy.nan.
tool_names	tuple[str, …]	Tool names, in the order they index the first axis.
metric_ids	tuple[str, …]	Metric ids, in the order they index the last axis. Every id resolves to a metric card.
dataset_names	tuple[str, …] | None	Dataset names indexing the middle axis when long, otherwise None.
layout	str	"wide" or "long".
source_path	str | None	The file the scores were read from, or None when constructed directly from arrays. Recorded in the run manifest.
blinding_sha256	str | None	The seal fingerprint when these scores were produced by beam.blind, otherwise None. When set, a run on these scores records the blinding in its manifest.