← Back to API reference

neuralbench.transforms.SimilaritySplit

pydantic model neuralbench.transforms.SimilaritySplit

Perform train/val/test split based on similarity of sentence events.

Depending on the type of stimulus event that is expected, the behavior is as follows:

• For Audio events, propagate sentence mapping to Word events, then chunk Audio events based on Word events.

• For Keystroke events, propagate sentence mapping to Keystroke events.

• For Sentence or Word events, directly apply the similarity-based split.

Parameters:

use_sklearn_split – If True, use sklearn’s train_test_split after computing clusters, rather than using SimilaritySplitter’s deterministic cluster assignment. NOTE: valid_random_state and test_random_state are ignored unless use_sklearn_split is True.

Fields:

• stim_event_type (Literal['Sentence', 'Word', 'Audio', 'Keystroke'])

• test_random_state (int)

• test_split_ratio (float)

• threshold (float)

• use_sklearn_split (bool)

• valid_random_state (int)

• valid_split_ratio (float)

field stim_event_type: Literal['Sentence', 'Word', 'Audio', 'Keystroke'] [Required]

field valid_split_ratio: float = 0.2

field test_split_ratio: float = 0.2

field valid_random_state: int = 33

field test_random_state: int = 33

field threshold: float = 0.2

field use_sklearn_split: bool = False

requirements: tp.ClassVar[tuple[str, ...]] = ()

← Back to API reference