Expand description
Embeddenator VSA - Vector Symbolic Architecture Core
This crate provides the foundational Vector Symbolic Architecture (VSA) operations for sparse ternary representations used in Embeddenator.
§Overview
VSA is a computational framework for representing and manipulating high-dimensional vectors through algebraic operations:
- Bundle (⊕): Superposition of multiple vectors
- Bind (⊙): Compositional binding of associations
- Similarity: Cosine similarity for pattern retrieval
§Core Types
SparseVec: Sparse ternary vector representationPackedTritVec: Memory-efficient packed trit storageCodebook: Mapping between symbols and vectors
§Example
use embeddenator_vsa::{SparseVec, ReversibleVSAConfig};
// Create random sparse vectors
let a = SparseVec::random();
let b = SparseVec::random();
// Bundle operation (superposition)
let bundled = a.bundle(&b);
// Compute similarity
let similarity = a.cosine(&b);
assert!(similarity >= -1.0 && similarity <= 1.0);Re-exports§
pub use codebook::BalancedTernaryWord;pub use codebook::Codebook;pub use codebook::CodebookTrainingConfig;pub use codebook::ProjectionConfig;pub use codebook::ProjectionResult;pub use codebook::SemanticOutlier;pub use codebook::WordMetadata;pub use dimensional::DifferentialEncoder;pub use dimensional::DifferentialEncoding;pub use dimensional::DimensionalConfig;pub use dimensional::HyperVec;pub use dimensional::Trit as DimTrit;pub use dimensional::TritDepthConfig;pub use dimensional::Tryte;pub use phase_training::train_codebook_with_phases;pub use phase_training::PhaseTrainer;pub use phase_training::PhaseTrainingConfig;pub use phase_training::PhaseTrainingStats;pub use phase_training::TrainingPhase;pub use reversible_encoding::ReversibleVSAEncoder;pub use reversible_encoding::MAX_POSITIONS;pub use simd_cosine::cosine_neon;pub use simd_cosine::cosine_scalar;pub use simd_cosine::cosine_simd;pub use ternary::CorrectionEntry;pub use ternary::ParityTrit;pub use ternary::Trit;pub use ternary::Tryte3;pub use ternary::Word6;pub use ternary_vec::PackedTritVec;pub use vsa::ReversibleVSAConfig;pub use vsa::SparseVec;pub use vsa::SparsityScaling;pub use vsa::VsaConfig;pub use vsa::VsaConfigSchema;pub use vsa::DIM;pub use coherency::CoherencyManager;pub use coherency::CoherencyState;pub use coherency::CoherencyStats;pub use coherency::CoherentEngram;pub use coherency::SyncProtocol;pub use coherency::Tier;pub use coherency::TierMask;pub use coherency::TieredBlock;pub use coherency::TieredCoherencyStats;pub use coherency::TieredState;pub use coherency::WritePolicy;pub use vram_pool::VramHandle;pub use vram_pool::VramPool;pub use vram_pool::VramPoolConfig;pub use vram_pool::VramPoolStats;pub use virtual_memory::MemoryTier;pub use virtual_memory::VMemHandle;pub use virtual_memory::VirtualMemory;pub use virtual_memory::VirtualMemoryConfig;pub use virtual_memory::VirtualMemoryError;pub use virtual_memory::VirtualMemoryStats;pub use resonator::FactorizationResult;pub use resonator::RecoveredFactor;pub use resonator::Resonator;pub use resonator::ResonatorConfig;pub use resonator::ResonatorStats;pub use resonator::SemanticInference;pub use resonator::TrainingExample;pub use resonator::TrainingResult;
Modules§
- codebook
- Codebook - Differential Encoding Base Model
- coherency
- Host-Device Coherency Protocol for GPU VRAM
- dimensional
- Dimensional Configuration - Variable Precision Hyperdimensional Substrate
- phase_
training - Deterministic Phase Training for Codebook Optimization
- resonator
- Resonator Networks for Learned Codebooks
- reversible_
encoding - Reversible Position-Aware VSA Encoding
- simd_
cosine - SIMD-accelerated cosine similarity for sparse ternary vectors
- ternary
- Foundational Balanced Ternary Primitives
- ternary_
vec - Packed ternary vector representation.
- virtual_
memory - Virtual Memory Abstraction for Tiered Storage
- vram_
pool - GPU VRAM Memory Pool for Persistent Engrams
- vsa
- Vector Symbolic Architecture (VSA) Implementation
Enums§
- VsaError
- Unified error type for VSA operations