Skip to main content

JAX + Units

Project description

JAX + Units

Built with JAX and Pint!

This module provides an interface between JAX and Pint to allow JAX to support operations with units. The propagation of units happens at trace time, so jitted functions should see no runtime cost. This library is experimental so expect some sharp edges.

For example:

>>> import jax
>>> import jax.numpy as jnp
>>> import jpu
>>>
>>> u = jpu.UnitRegistry()
>>>
>>> @jax.jit
... def add_two_lengths(a, b):
...     return a + b
...
>>> add_two_lengths(3 * u.m, jnp.array([4.5, 1.2, 3.9]) * u.cm)
<Quantity([3.045 3.012 3.039], 'meter')>

Installation

To install, use pip:

python -m pip install jpu

The only dependencies are jax and pint, and these will also be installed, if not already in your environment. Take a look at the JAX docs for more information about installing JAX on different systems.

Usage

Here is a slightly more complete example:

>>> import jax
>>> import numpy as np
>>> from jpu import UnitRegistry, numpy as jnpu
>>>
>>> u = UnitRegistry()
>>>
>>> @jax.jit
... def projectile_motion(v_init, theta, time, g=u.standard_gravity):
...     """Compute the motion of a projectile with support for units"""
...     x = v_init * time * jnpu.cos(theta)
...     y = v_init * time * jnpu.sin(theta) - 0.5 * g * jnpu.square(time)
...     return x.to(u.m), y.to(u.m)
...
>>> x, y = projectile_motion(
...     5.0 * u.km / u.h, 60 * u.deg, np.linspace(0, 1, 50) * u.s
... )

Technical details & limitations

The most significant limitation of this library is the fact that users must use jpu.numpy functions when interacting with "quantities" with units instead of the jax.numpy interface. This is because JAX does not (yet?) provide a general interface for dispatching of ufuncs on custom array classes. I have played around with the undocumented __jax_array__ interface, but it's not really flexible enough, and it isn't currently compatible with Pytree objects.

So far, only a subset of the numpy/jax.numpy interface is implemented. Pull requests adding broader support (including submodules) would be welcome!

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

jpu-0.0.1.tar.gz (15.6 kB view details)

Uploaded Source

Built Distribution

jpu-0.0.1-py3-none-any.whl (12.4 kB view details)

Uploaded Python 3

File details

Details for the file jpu-0.0.1.tar.gz.

File metadata

  • Download URL: jpu-0.0.1.tar.gz
  • Upload date:
  • Size: 15.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/4.0.2 CPython/3.11.4

File hashes

Hashes for jpu-0.0.1.tar.gz
Algorithm Hash digest
SHA256 3c02e3a6189d6f99c5d8188b349972907a28731d7db3924a6ea9eda31e83aabc
MD5 be9282196308a8623a67a6a1a50e82b9
BLAKE2b-256 e11b0d800e3e2ae58746418b1d4a5f0d4ddaae6ed4b096b0353748a0bee968e6

See more details on using hashes here.

File details

Details for the file jpu-0.0.1-py3-none-any.whl.

File metadata

  • Download URL: jpu-0.0.1-py3-none-any.whl
  • Upload date:
  • Size: 12.4 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/4.0.2 CPython/3.11.4

File hashes

Hashes for jpu-0.0.1-py3-none-any.whl
Algorithm Hash digest
SHA256 8427e18653aa711495617bd494a7da511f5b68d0f4dc6960b694126aebebd10e
MD5 6b726883a6b2e2fa9c4ff571096fa96e
BLAKE2b-256 c1144d312c93efbc87f29e99043b5ec1b3a42528d54d1edabec5024fd30285a0

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page