Configurations

This page provides a comprehensive reference for the configuration models used across different text-mallet algorithms.

POS Filter Configuration ("pos-filter")

Configures token masking based on specific Part-of-Speech tags.

POSFilterConfig Parameters

Parameter	Type	Default	Description
filter_type	FilterType or List[FilterType]	FilterType.Retain	Action to take on the target tags (e.g., retain them or remove them).
pos_tags	List[POSTag]	[NOUN, PROPN]	A collection of Universal POS tags targeted by the filter.
replacement_mechanism	str or List[str]	"default"	Strategy used to replace targeted text (e.g., replacing with the tag name).

—

Shannon Filter Configuration ("shannon")

Configures text filtering using information-theoretic metrics calculated from an underlying language model context.

ShannonFilterConfig Parameters

Parameter	Type	Default	Description
threshold	float or List[float]	10	The information surprisal ceiling or floor value used to trigger a mask.
bound	str or List[str]	"upper"	Determines if values above ("upper") or below ("lower") the threshold are masked.
replacement_mechanism	str or List[str]	"default"	Strategy used to replace targeted text.
max_context_length	int	8192	Max token historical window parsed by the LM for context evaluation.
output_mi_values	bool	False	If true, outputs calculated mutual information/surprisal values along with the text.

—

Linear Scramble Configuration ("scramble-BoW")

Configures a lightweight, framework-free Bag-of-Words shuffling operation.

LinearScrambleConfig Parameters

Parameter	Type	Default	Description
level	str or List[str]	"document"	The text layer boundaries applied to shuffles. Allowed values are typically "sentence" or "document".
seed	int	DEFAULT_SEED	Pseudorandom generator state seed to guarantee deterministic text layouts.

—

Hierarchical Scramble Configuration ("scramble-hier")

Configures deep syntactic structural scrambled layouts using dependency trees generated by a spaCy backend.

HierarchicalScrambleConfig Parameters

Parameter	Type	Default	Description
strength	str or List[str]	"strong"	Rearrangement severity settings. Typical options are "weak" (sibling shuffling) or "strong" (sibling shuffling + directional flips).
seed	int	DEFAULT_SEED	Pseudorandom generator seed utilized when scrambling leaf node distributions.

Multi-Obfuscation Feature: For parameters that accept a union type of a single value or a list (e.g., str | List[str]), passing a list of choices will cause TMallet to compute multiple obfuscation passes simultaneously. The result will be returned as a nested dictionary keyed by the options instead of a flat string. This is with the single exception of passing POS tags to the POS filter, as this must be an array and will not automatically provide the result as a dict. If you want to compute multuple obfuscation configurations at once, passing multiple parameters in the single config as an array leads to performance gains as the text is only processed a single time where possible.