Skip to content

Python Units Library Feature Matrix

A comprehensive comparison of Python units libraries to help you choose the right tool for your project.

Overview

Library Primary Focus Best For
dimtensor Physics + ML PyTorch/JAX workflows, GPU computing
Pint General purpose Flexible unit systems, data analysis
Astropy Astronomy Astronomical calculations, FITS files
unyt Astrophysics yt project, large simulations

Core Features

Feature dimtensor Pint Astropy unyt
NumPy arrays Yes Yes Yes Yes
SI units Yes Yes Yes Yes
CGS units Yes Yes Yes Yes
Imperial units Yes Yes Yes Yes
Custom units Limited Extensive Limited Limited
Unit simplification Auto Auto Auto Auto

Framework Integration

Feature dimtensor Pint Astropy unyt
PyTorch native Yes No No No
JAX native Yes No No No
Autograd support Yes No No No
GPU (CUDA) Yes No No No
GPU (MPS/Apple) Yes No No No
Dask arrays No Yes No Yes
Sparse arrays No Yes No No

Scientific Computing

Feature dimtensor Pint Astropy unyt
Physical constants CODATA 2022 scipy.constants Built-in yt constants
Uncertainty propagation Built-in Extension Separate No
Temperature handling Yes Yes Yes Yes
Logarithmic units (dB) No Yes Yes No
Equivalencies Basic Basic Advanced Basic

Domain-Specific Units

Domain dimtensor Pint Astropy unyt
Astronomy Module Basic Comprehensive Comprehensive
Chemistry Module Basic No No
Engineering Module Basic No No
Cosmology No No Yes Yes

I/O Support

Format dimtensor Pint Astropy unyt
JSON Yes Manual No Yes
HDF5 Yes Manual Table Yes
Parquet Yes No No No
NetCDF Yes No No No
pandas Yes pint-pandas No Limited
xarray Yes pint-xarray No No
FITS No No Native No
VOTable No No Native No

Performance

Metric dimtensor Pint Astropy unyt
Overhead vs NumPy 2-5x 2-5x 2-3x 1.5-3x
Memory efficiency Good Good Good Good
GPU acceleration Yes No No No
JIT compilation JAX No No No

Documentation & Community

Metric dimtensor Pint Astropy unyt
Documentation Good Excellent Excellent Good
Tutorials Growing Many Many Some
Stack Overflow New Active Active Moderate
GitHub stars Growing ~2k ~4k (main) ~400

Installation Size

Library Install Size Dependencies
dimtensor ~500 KB numpy
Pint ~1 MB numpy
Astropy ~50 MB numpy, scipy, etc.
unyt ~2 MB numpy, sympy

Code Examples

Creating Arrays

from dimtensor import DimArray, units
arr = DimArray([1, 2, 3], units.m / units.s)

import pint
ureg = pint.UnitRegistry()
arr = [1, 2, 3] * ureg.meter / ureg.second

from astropy import units as u
arr = [1, 2, 3] * u.m / u.s

from unyt import unyt_array
arr = unyt_array([1, 2, 3], 'm/s')

Unit Conversion

distance = DimArray([1000], units.m)
in_km = distance.to(units.km)  # 1 km

distance = 1000 * ureg.meter
in_km = distance.to('km')  # 1 km

distance = 1000 * u.m
in_km = distance.to(u.km)  # 1 km

distance = unyt_array([1000], 'm')
in_km = distance.to('km')  # 1 km

Decision Guide

Choose dimtensor if:

  • You need PyTorch or JAX integration
  • You're doing machine learning with physical units
  • You need GPU acceleration
  • You want built-in uncertainty propagation
  • You need multiple I/O formats

Choose Pint if:

  • You need maximum flexibility in unit definitions
  • You're doing pure data analysis (no ML)
  • You have complex custom unit systems
  • You need pandas integration

Choose Astropy if:

  • You're doing astronomy/astrophysics
  • You need spectral equivalencies
  • You work with FITS files
  • You use other Astropy tools

Choose unyt if:

  • You use the yt project
  • You're doing astrophysical simulations
  • You need cosmological units
  • You work with very large datasets (Dask)

Summary

For machine learning and GPU computing: dimtensor

For general purpose flexibility: Pint

For astronomy: Astropy

For yt/simulations: unyt