Available Distributions
All distributions live in sage.data.waveform.distributions and are referenced
by name in gwconfig.yaml. Every distribution exposes a sample(shape, device,
dtype, generator) interface that runs entirely on the target device.
uniform
Class: Uniform
Draws samples uniformly from a closed interval [min, max].
Used for masses, spin magnitudes, coalescence time, and GPS injection time.
mass1:
name: uniform
min: 7.0
max: 50.0
Physics: A uniform mass prior is maximally agnostic about the mass distribution. It overrepresents high masses relative to the astrophysical population but ensures the network is exposed to the full range without bias toward any mass scale.
uniform_angle
Class: UniformAngle
Draws angles uniformly from [0, 2π) with optional wrapping. Used for coalescence
phase and polarisation angle, which are both cyclic.
coa_phase:
name: uniform_angle
polarization:
name: uniform_angle
Physics: The coalescence phase and polarisation angle have no preferred value for a population of unresolved sources, so a uniform prior over the full cycle is the natural choice.
sin_angle
Class: SinAngle
Draws polar angles from the distribution proportional to sin(θ) over [0, π]
via the inverse-CDF method in cos(θ) space. Used for the binary inclination angle.
inclination:
name: sin_angle
Physics: A uniform distribution over a sphere has a polar-angle marginal
proportional to sin(θ). Using this prior for inclination ensures that face-on and
edge-on orientations occur with their correct relative frequency for a population of
isotropically oriented binaries.
uniform_sky
Class: UniformSky
Jointly samples right ascension and declination so that the resulting sky positions are
uniform over the 2-sphere. RA is drawn uniformly from [0, 2π); declination is drawn
from a cos(δ)-weighted distribution via CosAngle.
sky:
name: uniform_sky
ra: ra
dec: dec
Physics: An isotropic source population has sky positions that are uniform in solid
angle, which means uniform in RA and uniform in sin(dec). Naïvely drawing dec from
a flat distribution would oversample the poles.
uniform_solidangle
Class: UniformSolidAngle
Jointly samples a polar angle (via sin_angle) and an azimuthal angle
(via uniform_angle) to produce orientation vectors that are isotropically
distributed on the sphere. Used for 3-D spin orientations.
spin1_orientation:
name: uniform_solidangle
polar-angle: spin1_polar
azimuthal-angle: spin1_azimuthal
spin2_orientation:
name: uniform_solidangle
polar-angle: spin2_polar
azimuthal-angle: spin2_azimuthal
Physics: Spin orientations in the absence of tidal alignment are expected to be isotropically distributed. Uniform solid-angle sampling (rather than independent uniform in both angles) correctly implements this.
uniform_radius
Class: UniformRadius
Draws distances from a distribution proportional to r² over [min, max], which
corresponds to uniform spatial number density in a Euclidean volume. Used for chirp
distance as a proxy for luminosity distance.
chirp_distance:
name: uniform_radius
min: 130.0
max: 350.0
Physics: For a volume-limited survey of uniformly distributed sources, the number
of sources within distance r grows as r³, so p(r) ∝ r². Sampling chirp
distance rather than luminosity distance flattens the effective SNR distribution,
giving more training examples near the detection threshold.