dimtensor vs Astropy Units¶

Astropy provides a comprehensive units package designed for astronomy. This page compares it with dimtensor.

TL;DR¶

Feature	dimtensor	Astropy
Focus	General physics + ML	Astronomy
PyTorch support	Native	None
JAX support	Native	None
Astronomy units	Domain module	Comprehensive
Equivalencies	Basic	Advanced (spectral, etc.)
FITS I/O	No	Native
Uncertainty	Built-in	Separate NDData

When to Choose dimtensor¶

Choose dimtensor if you need:

Machine learning with units (PyTorch, JAX)
GPU acceleration
Built-in uncertainty propagation
Multi-framework support (NumPy + PyTorch + JAX)
General physics beyond astronomy

# dimtensor: ML with astronomy units
from dimtensor.torch import DimTensor
from dimtensor.domains.astronomy import solar_mass, parsec
import torch

mass = DimTensor(torch.tensor([1.0], requires_grad=True), solar_mass)
# Use in neural network training...

When to Choose Astropy¶

Choose Astropy if you need:

Comprehensive astronomy unit system
Spectral equivalencies (wavelength/frequency/energy)
FITS file I/O with units
Coordinate transformations
Integration with other Astropy tools

# Astropy: Spectral equivalencies
from astropy import units as u

wavelength = 500 * u.nm
frequency = wavelength.to(u.Hz, equivalencies=u.spectral())
energy = wavelength.to(u.eV, equivalencies=u.spectral())

Feature Comparison¶

Unit Systems¶

dimtensorAstropy

from dimtensor import DimArray, units
from dimtensor.domains.astronomy import parsec, AU, solar_mass, light_year

# Core astronomy units available
distance = DimArray([4.24], light_year)
distance_pc = distance.to(parsec)

from astropy import units as u

# Extensive astronomy unit system
distance = 4.24 * u.lyr
distance_pc = distance.to(u.pc)

# Specialized astronomy units
flux = 1e-23 * u.erg / u.s / u.cm**2 / u.Hz  # Jansky-like
magnitude = 5 * u.mag

Equivalencies¶

Astropy excels at equivalencies (conversions between physically related but dimensionally different quantities):

dimtensorAstropy

# dimtensor focuses on dimensional correctness
# Explicit conversions required
from dimtensor import DimArray, units
from dimtensor import constants

wavelength = DimArray([500e-9], units.m)  # 500 nm
frequency = constants.c / wavelength  # Manual conversion

from astropy import units as u

# Automatic equivalencies
wavelength = 500 * u.nm

# Convert between wavelength, frequency, and energy
freq = wavelength.to(u.Hz, equivalencies=u.spectral())
energy = wavelength.to(u.eV, equivalencies=u.spectral())

# Temperature equivalencies
temp = 5000 * u.K
wavelength_peak = temp.to(u.um, equivalencies=u.temperature_energy())

Machine Learning¶

dimtensorAstropy

import jax
from dimtensor.jax import DimArray
from dimtensor.domains.astronomy import solar_mass

@jax.jit
def gravitational_energy(m1, m2, r):
    from dimtensor import constants
    return -constants.G * m1 * m2 / r

# JIT-compiled, units preserved throughout

# Astropy units don't work with JAX or PyTorch
# Must strip units before ML operations
from astropy import units as u

mass = 1.0 * u.solMass
mass_value = mass.to(u.kg).value  # Strip units for ML

I/O Support¶

Format	dimtensor	Astropy
JSON	Yes	No
HDF5	Yes	Via Astropy Table
Parquet	Yes	No
NetCDF	Yes	No
FITS	No	Native
VOTable	No	Native

Uncertainty Handling¶

dimtensorAstropy

from dimtensor import DimArray
from dimtensor.domains.astronomy import parsec

# Built-in uncertainty
distance = DimArray([10.0], parsec, uncertainty=[0.5])
luminosity = distance ** 2  # Uncertainty propagates

from astropy import units as u
from astropy.nddata import NDDataArray, StdDevUncertainty

# Separate uncertainty container
data = NDDataArray([10.0] * u.pc,
                   uncertainty=StdDevUncertainty([0.5] * u.pc))

Migration from Astropy¶

If you're adding ML to astronomy code:

# Astropy code
from astropy import units as u
from astropy import constants as const

mass = 1.0 * u.solMass
velocity = 200 * u.km / u.s
kinetic_energy = 0.5 * mass * velocity**2

# dimtensor equivalent (for ML compatibility)
from dimtensor import DimArray
from dimtensor.domains.astronomy import solar_mass
from dimtensor import units

mass = DimArray([1.0], solar_mass)
velocity = DimArray([200], units.km / units.s)
kinetic_energy = 0.5 * mass * velocity**2
# Now can use with PyTorch/JAX

Using Both Together¶

For astronomy projects needing ML, consider using both:

from astropy import units as u
from dimtensor import DimArray, units as dt_units
from dimtensor.torch import DimTensor
import torch

# Load data with Astropy (FITS, coordinates, etc.)
from astropy.io import fits
data = fits.getdata('observations.fits')
astropy_flux = data * u.Jy

# Convert to dimtensor for ML
flux_values = astropy_flux.to(u.W / u.m**2 / u.Hz).value
flux = DimTensor(torch.tensor(flux_values), dt_units.W / dt_units.m**2 / dt_units.Hz)

# Train your model with units preserved...

Conclusion¶

Use dimtensor for ML workflows, GPU computing, or general physics
Use Astropy for astronomy-specific workflows, FITS files, or coordinate transformations
Use both when you need Astropy's astronomy features plus ML capabilities