Crate tsetlin_rs 

Tsetlin Machine

A professional Rust implementation of the Tsetlin Machine algorithm for interpretable machine learning.

Models

• TsetlinMachine - Binary classification (dynamic)
• SmallTsetlinMachine - Binary classification (compile-time optimized)
• SparseTsetlinMachine - Memory-efficient inference (5-100x reduction)
• MultiClass - Multi-class classification (one-vs-all)
• Regressor - Regression
• Convolutional - Image classification with patch extraction

Clause Types

Type	Heap	Best For
Clause	Yes	Dynamic dimensions, serde
SmallClause	No	N ≤ 64, maximum speed
BitwiseClause	Yes	N ≥ 64, 25-92x speedup
SmallBitwiseClause	No	64-256 features, no heap
SparseClause	Inline	Inference, 5-100x compression
SparseClauseBank	Yes	CSR batch inference

Advanced Features

• Weighted Clauses - Clauses learn weights based on accuracy
• Adaptive Threshold - Dynamic T adjustment during training
• Clause Pruning - Automatic reset of dead/ineffective clauses
• Const Generics - Zero-allocation stack types with loop unrolling
• Lock-Free Parallel Training - Async local voting tallies (ICML 2021)

Feature Flags

• std (default): Standard library support
• simd: SIMD-optimized evaluation (requires nightly)
• parallel: Parallel training via rayon
• serde: Serialization support

Quick Start

use tsetlin_rs::{Config, TsetlinMachine};

let config = Config::builder().clauses(20).features(2).build().unwrap();
let mut tm = TsetlinMachine::new(config, 10);

let x = vec![vec![0, 0], vec![0, 1], vec![1, 0], vec![1, 1]];
let y = vec![0, 1, 1, 0];

tm.fit(&x, &y, 200, 42);
assert!(tm.evaluate(&x, &y) >= 0.75);

Re-exports

pub use error::Error;

pub use error::Result;

Modules

error

Error types for Tsetlin Machine.

feedback

Type I and Type II feedback mechanisms.

utils

Utility functions for random number generation and helpers.

Structs

AdvancedOptions

Overview

Automaton

Overview

BitPlaneBank

Bit-plane storage for automata states.

BitwiseClause

Overview

Clause

A clause with 2*n_features automata, weight, and activation tracking.

ClauseBank

Structure of Arrays storage for Tsetlin Machine clauses.

ClauseFilter

Bloom filter for fast clause rejection.

ClauseFilterStats

Statistics about a clause filter.

Config

Overview

ConfigBuilder

Overview

ConvConfig

Overview

Convolutional

Overview

EarlyStop

Early stopping configuration.

FitOptions

Options for training a Tsetlin Machine.

FitResult

Result of training.

MultiClass

Multi-class Tsetlin Machine using one-vs-all strategy.

Regressor

Overview

Rule

Overview

SmallBitwiseClause

Bitwise clause with compile-time known feature count.

SmallClause

A clause with compile-time known feature count.

SmallTsetlinMachine

Binary classification Tsetlin Machine with compile-time known dimensions.

SparseClause

Sparse clause representation storing only active literal indices.

SparseClauseBank

Sparse clause bank using CSR (Compressed Sparse Row) format.

SparseMemoryStats

Memory usage breakdown for sparse clause bank.

SparseTsetlinMachine

Sparse Tsetlin Machine for memory-efficient inference.

TsetlinMachine

Overview

Traits

TsetlinModel

Unified interface for all Tsetlin Machine variants.

VotingModel

Extension trait for models with vote-based predictions.

Functions

pack_batch

Overview

pack_input

Overview

pack_input_small

Packs binary input into u64 array for bitwise evaluation.

Type Aliases

BitwiseClause64

Bitwise clause for 64 features (1 u64 word).

BitwiseClause128

Bitwise clause for 128 features (2 u64 words).

BitwiseClause256

Bitwise clause for 256 features (4 u64 words).

Clause2

Type alias for 2-feature clause.

Clause4

Type alias for 4-feature clause (XOR problems).

Clause8

Type alias for 8-feature clause.

Clause16

Type alias for 16-feature clause.

Clause32

Type alias for 32-feature clause.

Clause64

Type alias for 64-feature clause.

TM2x20

2-feature, 20-clause machine (XOR).

TM4x40

4-feature, 40-clause machine.

TM8x80

8-feature, 80-clause machine.

TM16x160

16-feature, 160-clause machine.