HoloVec Documentation¶

Brain-Inspired Computing with Hyperdimensional Vectors

A high-performance Python library for building cognitive AI systems using vector symbolic architectures. Encode any data type into high-dimensional space for fast, robust, and explainable machine learning.

Get Started Browse Examples

Why HoloVec?¶

🧠 Brain-Inspired

Hyperdimensional computing mimics how the brain represents and processes information using high-dimensional distributed representations.

Bio-plausible operations that map naturally to neural circuits.

⚡ Fast & Efficient

Vectorized operations on 10,000-dimensional vectors are surprisingly efficient. No neural network training required.

Process data in real-time with minimal computational overhead.

🛡️ Robust to Noise

Gracefully handles noise, partial corruption, and approximate data. Ideal for edge devices and fault-tolerant systems.

Performance degrades gracefully under hardware errors.

🔍 Explainable

Operations are transparent and interpretable. You can visualize and understand what the model is doing.

No black box. Every operation has clear semantics.

🎓 One-Shot Learning

Learn from very few examples. No gradient descent required. Perfect for online learning and rapid adaptation.

Add new knowledge incrementally without retraining.

📦 Versatile

Works with any data type - text, images, sensors, symbolic structures. Perfect for multimodal fusion.

Unify different modalities in the same representational space.

What You Can Build¶

📝 Text & NLP

Sentiment analysis, document classification, chatbot intent recognition, semantic search

🖼️ Computer Vision

Real-time image recognition, gesture detection, visual similarity search, anomaly detection

🎯 Recommender Systems

Personalized recommendations, content filtering, collaborative filtering with explainable results

📊 Sensor & IoT

Time-series classification, sensor fusion, predictive maintenance, edge device inference

🔗 Multimodal AI

Text + image fusion, audio-visual processing, cross-modal retrieval and reasoning

🧪 Research & Prototyping

Fast experimentation, one-shot learning tasks, cognitive modeling, neurosymbolic AI

HoloVec Features¶

A comprehensive library for hyperdimensional computing with everything you need from research to production.

Quick Start¶

Installation

Install holovec using uv (recommended) or pip:

# Using uv (recommended)
uv pip install holovec

# Or using pip
pip install holovec

Optional Backends

For GPU acceleration or specialized hardware, install backend-specific dependencies:

# GPU support with PyTorch
uv pip install holovec[torch]

# JIT compilation and TPU support with JAX
uv pip install holovec[jax]

Example Usage¶

Explore how HoloVec handles different types of data and use cases. Each example demonstrates complete workflows from encoding to querying.

Symbolic Reasoning

from holovec import VSA

# Create model
model = VSA.create('FHRR', dim=10000, seed=42)

# Encode symbols
alice = model.random(seed=1)
bob = model.random(seed=2)
loves = model.random(seed=3)

# Create "Alice loves Bob"
statement = model.bind(model.bind(alice, loves), bob)

# Query: Who does Alice love?
query = model.unbind(model.unbind(statement, alice), loves)
print(f"Similarity to Bob: {model.similarity(query, bob):.3f}")
# Output: 1.0

Continuous Values

from holovec import VSA
from holovec.encoders import FractionalPowerEncoder

# Create model
model = VSA.create('FHRR', dim=10000, seed=42)

# Create encoder for temperatures (0-100°C)
temp_encoder = FractionalPowerEncoder(
    model,
    min_val=0,
    max_val=100,
    bandwidth=0.1
)

# Encode temperatures
temp_25 = temp_encoder.encode(25.0)
temp_26 = temp_encoder.encode(26.0)
temp_50 = temp_encoder.encode(50.0)

# Similar temperatures have high similarity
print(f"25°C vs 26°C: {model.similarity(temp_25, temp_26):.3f}")
# Output: 0.95

print(f"25°C vs 50°C: {model.similarity(temp_25, temp_50):.3f}")
# Output: 0.32

Text Classification

from holovec import VSA
from holovec.encoders import NGramEncoder

# Create model
model = VSA.create('FHRR', dim=10000, seed=42)

# Create text encoder
text_encoder = NGramEncoder(model, n=3)

# Encode training documents
doc_positive = text_encoder.encode("I love this product!")
doc_negative = text_encoder.encode("This is terrible")

# Encode new review
new_review = text_encoder.encode("I really love it")

# Classify by similarity
sim_pos = model.similarity(new_review, doc_positive)
sim_neg = model.similarity(new_review, doc_negative)

print(f"Positive: {sim_pos:.3f}, Negative: {sim_neg:.3f}")
# Output: Positive: 0.87, Negative: 0.12