blind

blind(scores, seed=0, label_prefix=_DEFAULT_PREFIX)

Relabel and shuffle the tools of a score table, returning a seal.

The tool axis is permuted under seed and renamed to opaque labels such as method_1, method_2, so neither the names nor the row order carry the methods’ identity. The metric and dataset axes are left unchanged, since the analyst needs them to set polarity, weights and the cross-dataset rule. The returned scores carry the seal fingerprint so a run on them records the blinding in its manifest.

Parameters

Name	Type	Description	Default
scores	Scores	The score table or tensor to blind.	required
seed	int	Seed for the row permutation. The same seed reproduces the same blinding.	0
label_prefix	str	Prefix for the opaque labels. Default "method".	_DEFAULT_PREFIX

Returns

Type	Description
tuple of (Scores, Seal)	The blinded scores and the seal that unblinds them.

Examples

>>> import numpy as np
>>> from beam import Scores, blind, unblind
>>> s = Scores(
...     values=np.array([[0.9, 0.1], [0.5, 0.5]]),
...     tool_names=("seurat", "sc3"),
...     metric_ids=("ari", "runtime"),
...     dataset_names=None,
...     layout="wide",
... )
>>> blinded, seal = blind(s, seed=1)
>>> set(blinded.tool_names) == {"method_1", "method_2"}
True
>>> set(seal.mapping.values()) == {"seurat", "sc3"}
True
>>> tuple(unblind(blinded, seal).tool_names) == tuple(seal.translate(blinded.tool_names))
True