Vector Symbolic Architectures for compositional, high-dimensional computing.

Documentation | Installation | Quick Start | Examples

---

What HoloVec Is

HoloVec is a Python library for hyperdimensional computing (HDC) and vector symbolic architectures (VSA). It gives you a consistent API for:

• binding and unbinding structured representations
• bundling sets, prototypes, and memories
• permutation-based order encoding
• scalar, sequence, spatial, and structured encoders
• associative retrieval and cleanup memories

The library currently treats NumPy as the release-blocking backend. PyTorch and JAX are supported as optional backends when their environment-specific tests are enabled.

---

Installation

pip install holovec

Optional extras:

pip install "holovec[torch]"  # PyTorch backend
pip install "holovec[jax]"    # JAX backend
pip install "holovec[all]"    # torch + jax + dev + docs extras

For source work:

git clone https://github.com/Twistient/HoloVec.git
cd HoloVec
uv sync --extra dev

Use uv pip install -e .[dev] only when you explicitly want an editable install outside the normal uv sync workflow.

---

Quick Start

from holovec import VSA
from holovec.encoders import FractionalPowerEncoder
from holovec.retrieval import Codebook, ItemStore

model = VSA.create("FHRR", dim=4096, seed=7)

role_color = model.random(seed=1)
role_temp = model.random(seed=2)
red = model.random(seed=10)
apple = model.random(seed=11)

temp_encoder = FractionalPowerEncoder(
    model,
    min_val=0.0,
    max_val=100.0,
    bandwidth=1.5,
    seed=3,
)

record = model.bundle(
    [
        model.bind(role_color, red),
        model.bind(role_temp, temp_encoder.encode(24.0)),
    ]
)

recovered_color = model.unbind(record, role_color)
store = ItemStore(model).fit(
    Codebook({"red": red, "apple": apple}, backend=model.backend)
)

print(store.query(recovered_color, k=1))

If you want a runnable version of that workflow, start with examples/00_quickstart.py.

---

Choosing a Model

Model	Space	Inverse Style	Order Sensitivity	Typical Use
FHRR	Complex	Exact	No	General-purpose default, continuous encoders
GHRR	Matrix	Exact	Yes	Order-sensitive and nested structures
MAP	Bipolar	Self-inverse	No	Fast algebra, hardware-friendly workflows
HRR	Real/Bipolar	Approximate	No	Classic baseline and literature reproduction
VTB	Matrix/Real	Approximate	Yes	Directional bindings and asymmetric relations
BSC	Binary	Self-inverse	No	Binary and FPGA-style workflows
BSDC	Sparse	Approximate	No	Sparse memory-efficient representations
BSDC-SEG	Sparse segment	Self-inverse	No	Segment-sparse retrieval and fast pattern search

Start with FHRR unless you have a clear reason to prefer order-sensitive matrix models, self-inverse discrete models, or sparse segment-based storage.

Examples of model-specific factory configuration:

VSA.create("BSDC", dim=20000, sparsity=0.01, binding_mode="cdt")
VSA.create("BSDC-SEG", dim=400, segments=20)
VSA.create("GHRR", dim=96, matrix_size=3, diagonality=0.4)
VSA.create("VTB", dim=512, n_bases=4, temperature=50.0)
VSA.create("FHRR", dim=4096, backend="torch", device="cuda")

See the full comparison in docs/models/index.md.

---

Maintained Examples

These examples are maintained alongside the core API and exercised in automated smoke tests:

• examples/00_quickstart.py: create a model, encode data, bind, retrieve
• examples/02_models_comparison.py: compare model families
• examples/10_encoders_scalar.py: scalar encoders and decoding
• examples/13_encoders_position_binding.py: sequence encoding and decoding
• examples/26_retrieval_basics.py: codebooks, item stores, threshold retrieval, persistence
• examples/27_cleanup_strategies.py: brute-force vs resonator cleanup
• examples/41_model_ghrr_diagonality.py: GHRR order sensitivity
• examples/42_model_bsdc_seg.py: BSDC-SEG segment patterns

Additional examples remain in the repository, but the list above is the maintained starting path through the library.

---

Documentation Highlights

• Quick Start
• Model Comparison
• Design Patterns
• Performance Guidance
• Benchmark Methodology
• Migration Notes

---

Testing

uv run --extra dev pytest --no-cov
uv run --extra dev ruff check holovec tests
uv run --extra dev mypy holovec
uv sync --extra docs && uv run --extra docs mkdocs build --strict

The maintained example scripts are also exercised in automated smoke tests.

---

References

HoloVec implements models and encoders drawn from the HDC/VSA literature. For a fuller bibliography, see the documentation references.

Good starting points:

• Pentti Kanerva, Hyperdimensional Computing: An Introduction to Computing in Distributed Representation with High-Dimensional Random Vectors (2009)
• Tony Plate, Holographic Reduced Representation (2003)
• Schlegel, Neubert, and Protzel, A Comparison of Vector Symbolic Architectures (2022)
• Kleyko et al., A Survey on Hyperdimensional Computing / Vector Symbolic Architectures Parts I and II (2022, 2023)

---

Citation

For machine-readable citation metadata, see CITATION.cff.

@software{holovec_2026,
  author  = {Schroeder, Brodie},
  title   = {HoloVec: Vector Symbolic Architectures for Python},
  year    = {2026},
  version = {1.0.2},
  url     = {https://github.com/Twistient/HoloVec},
  license = {Apache-2.0}
}

---

Contributing

See CONTRIBUTING.md. Key areas: new models, encoders, documentation, and performance.

License

Apache 2.0. See LICENSE.

Contact

GitHub Issues • Discussions • brodie@twistient.com