msgpack-numpy-opentensor

What it is and what it does

msgpack-numpy bridges NumPy and msgpack by adding type-aware serialization for NumPy arrays and Python complex numbers. It works by providing custom encoder and decoder functions that preserve dtype information during msgpack serialization, or by monkey-patching msgpack globally so all serialization calls become NumPy-aware automatically.

The package is designed for scenarios where you need efficient, compact binary serialization of numerical data while preserving type information across serialization boundaries. It trades some storage overhead (for type metadata) against the convenience of not having to manually reconstruct dtypes after deserialization. The main constraint is that deserialized arrays are read-only, and arrays with object dtype ('O') fall back to pickle, which defeats much of msgpack's efficiency advantage.

Use it for:

• Serialize NumPy arrays to disk or network in a compact binary format while preserving dtype information.
• Make msgpack-based RPC or message-passing systems NumPy-aware without modifying existing msgpack call sites.
• Exchange numerical data between Python processes or services using msgpack as the transport layer.
• Store machine learning model weights or intermediate tensors in a format more efficient than JSON or pickle.
• Integrate NumPy serialization into existing msgpack-based data pipelines without rewriting encoder/decoder logic.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Serializes and deserializes NumPy arrays and Python complex types using msgpack's efficient binary format, with optional monkey-patching to make msgpack globally NumPy-aware.

Yes, if you already use msgpack and need NumPy array serialization with type preservation. No, if you are starting a new project—the dormant maintenance (last release 2023-10-02, no recent commits) and low GitHub stars (2) suggest limited community support and risk of compatibility drift with future NumPy or msgpack releases. Consider pickle or Protocol Buffers for new work unless msgpack integration is a hard requirement.

Install

msgpack-numpy-opentensor on PyPI

pip install msgpack-numpy-opentensor

uv add msgpack-numpy-opentensor

poetry add msgpack-numpy-opentensor

Installing msgpack-numpy-opentensor

Before you install

Low friction: pure wheel distribution with only two runtime dependencies (numpy and msgpack). Maintenance is dormant—last release was 2023-10-02 and no commits since then—so expect no active bug fixes or updates.

License in practice

BSD permissive license allows use in most commercial and open-source projects without significant restrictions.

Quickstart

import msgpack
import msgpack_numpy as m
import numpy as np

m.patch()  # Make all msgpack calls NumPy-aware
x = np.random.rand(5)
x_enc = msgpack.packb(x)
x_rec = msgpack.unpackb(x_enc)

Deserialized NumPy arrays are read-only; copy them before modification. Arrays with dtype 'O' fall back to pickle, negating msgpack efficiency gains.

Verify before relying

• Whether the package works reliably with modern NumPy and msgpack versions beyond what classifiers claim.
• Performance characteristics compared to alternatives like pickle or Protocol Buffers for typical workloads.
• Whether the dormant maintenance status poses compatibility risks with future Python releases.

Package facts

Evidence: msgpack_numpy_opentensor-0.5.0-py2.py3-none-any.whl

Tags

More Software Development packages