fast_lisa_subtraction.priors.analytical module

Some standard distributions for sampling the priors.

class fast_lisa_subtraction.priors.analytical.Cauchy(loc, scale, minimum=None, maximum=None, name=None, device='cpu')[source]

Bases: Prior

Cauchy prior distribution.

Parameters:

loc (float) – Location parameter of the Cauchy prior.
scale (float) – Scale parameter of the Cauchy prior.
minimum (float or None, optional) – Lower bound of the support.
maximum (float or None, optional) – Upper bound of the support.
name (str, optional) – Name of the prior parameter.
device (str, optional) – Torch device used for sampling.

sample(num_samples, standardize=False)[source]

Sample from the Cauchy prior.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior.

Return type:

torch.Tensor

class fast_lisa_subtraction.priors.analytical.Gamma(alpha, beta, offset=0, minimum=None, maximum=None, name=None, device='cpu')[source]

Bases: Prior

Gamma prior distribution with an optional location shift.

Parameters:

alpha (float) – Shape parameter of the Gamma prior.
beta (float) – Rate parameter of the Gamma prior (inverse scale).
offset (float, optional) – Additive offset applied to samples.
minimum (float or None, optional) – Lower bound of the support.
maximum (float or None, optional) – Upper bound of the support.
name (str, optional) – Name of the prior parameter.
device (str, optional) – Torch device used for sampling.

Notes

The unshifted Gamma density is

\[p(\theta) = \frac{\beta^{\alpha}}{\Gamma(\alpha)} \theta^{\alpha - 1} e^{-\beta \theta}.\]

Samples are drawn from torch.distributions.Gamma(alpha, beta) and then shifted by a median-based centering m and the provided offset:

\[x = z - m + \mathrm{offset},\]

where z is Gamma-distributed and m is computed from the inverse incomplete gamma function.

log_prob(x, standardize=False)[source]

Evaluate the log-probability of the shifted Gamma prior.

Parameters:

x (array-like or torch.Tensor) – Points at which to evaluate log p(x).
standardize (bool, optional) – If True, interpret x as standardized and de-standardize before evaluating.

Returns:

Log-probability values.

Return type:

torch.Tensor

Notes

The evaluation uses the shifted variable

\[z = x - \mathrm{offset} + m,\]

and returns -inf where z <= 0.

sample(num_samples, standardize=False)[source]

Sample from the shifted Gamma prior.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior.

Return type:

torch.Tensor

class fast_lisa_subtraction.priors.analytical.Gaussian(mean, std_dev, name=None, device='cpu')[source]

Bases: Prior

Gaussian prior distribution.

Parameters:

mean (float) – Mean of the Gaussian prior.
std_dev (float) – Standard deviation of the Gaussian prior.
name (str, optional) – Name of the prior parameter.
device (str, optional) – Torch device used for sampling.

sample(num_samples, standardize=False)[source]

Sample from the Gaussian prior.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior.

Return type:

torch.Tensor

class fast_lisa_subtraction.priors.analytical.LogNormal(mu, sigma, minimum=None, maximum=None, name=None, device='cpu')[source]

Bases: Prior

Shifted log-normal prior distribution.

Parameters:

mu (float) – Location parameter used as an additive shift.
sigma (float) – Log-space standard deviation.
minimum (float or None, optional) – Lower bound of the support.
maximum (float or None, optional) – Upper bound of the support.
name (str, optional) – Name of the prior parameter.
device (str, optional) – Torch device used for sampling.

Notes

Sampling draws y from a log-normal distribution and returns a shifted value:

\[x = y + \mu, \quad y \sim \mathrm{LogNormal}(0, \sigma).\]

log_prob(x, standardize=False)[source]

Evaluate the log-probability of the shifted log-normal prior.

Parameters:

x (array-like or torch.Tensor) – Points at which to evaluate log p(x).
standardize (bool, optional) – If True, interpret x as standardized and de-standardize before evaluating.

Returns:

Log-probability values.

Return type:

torch.Tensor

Notes

The evaluation uses the shifted variable

\[y = x - \mu,\]

and applies a log-normal density with parameters (mu, sigma) to y.

sample(num_samples, standardize=False)[source]

Sample from the shifted log-normal prior.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior.

Return type:

torch.Tensor

class fast_lisa_subtraction.priors.analytical.MultivariatePrior(priors)[source]

Bases: object

Container for a multivariate distribution built from 1D priors.

Parameters:: priors (dict or list) – Collection of Prior instances. If a dict is provided, the values are used in the order of the keys.

destandardize(samples)[source]

Reverse standardization using component prior statistics.

Parameters:: samples (dict) – Mapping from parameter name to standardized samples.
Returns:: De-standardized samples.
Return type:: dict

property maximums

Maximum bounds of the component priors.

Returns:: Mapping from parameter name to maximum value.
Return type:: dict

property means

Means of the component priors.

Returns:: Mapping from parameter name to mean.
Return type:: dict

property metadata

Metadata associated with the multivariate prior.

Returns:

Dictionary containing:

meansdict: Mapping from parameter name to mean.
stdsdict: Mapping from parameter name to standard deviation.
boundsdict: Mapping from parameter name to (min, max).
inference_parameterslist of str: Parameter names in order.

Return type:

dict

property minimums

Minimum bounds of the component priors.

Returns:: Mapping from parameter name to minimum value.
Return type:: dict

property names

Names of the component priors.

Returns:: Names of each prior in order.
Return type:: list of str

sample(num_samples, standardize=False)[source]

Sample from each component prior.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior.

Return type:

TensorSamples

sample_array(num_samples, standardize=False)[source]

Sample from the prior distribution and return a NumPy array.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior as an array of shape (N, D).

Return type:

numpy.ndarray

standardize(samples)[source]

Standardize samples using component prior statistics.

Parameters:: samples (dict) – Mapping from parameter name to samples.
Returns:: Standardized samples.
Return type:: dict

property stds

Standard deviations of the component priors.

Returns:: Mapping from parameter name to standard deviation.
Return type:: dict

to_array(samples)[source]

Convert a dictionary of samples to a 2D NumPy array.

Parameters:: samples (dict) – Mapping from parameter name to samples (torch tensors).
Returns:: Array of shape (N, D) where D is the number of priors.
Return type:: numpy.ndarray

to_dict(samples)[source]

Convert a 2D array of samples to a dictionary.

Parameters:: samples (numpy.ndarray) – Array of shape (N, D).
Returns:: Mapping from parameter name to column arrays.
Return type:: dict

class fast_lisa_subtraction.priors.analytical.Prior(minimum=None, maximum=None, name='Parameter', device='cpu')[source]

Bases: object

Base class for analytical priors.

Parameters:

minimum (float or None, optional) – Lower bound of the prior support. If None, the bound is estimated from sampled values.
maximum (float or None, optional) – Upper bound of the prior support. If None, the bound is estimated from sampled values.
name (str, optional) – Parameter name.
device (str, optional) – Torch device used for sampling and computations.

clip(samples)[source]

Filter samples to the prior support.

Parameters:: samples (torch.Tensor) – Samples to clip.
Returns:: Samples within [minimum, maximum].
Return type:: torch.Tensor

destandardize(samples)[source]

Reverse standardization to the original scale.

Parameters:: samples (torch.Tensor) – Standardized samples.
Returns:: Samples in the original scale.
Return type:: torch.Tensor

property maximum

Upper bound of the prior support.

Returns:: Maximum value of the support.
Return type:: float or torch.Tensor

property mean

Mean of the prior.

Returns:: Mean estimated from sampled values.
Return type:: torch.Tensor

property minimum

Lower bound of the prior support.

Returns:: Minimum value of the support.
Return type:: float or torch.Tensor

property name

Parameter name.

Returns:: Name of the parameter.
Return type:: str

sample(num_samples, standardize=False)[source]

Draw samples from the prior.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior.

Return type:

torch.Tensor

Raises:

NotImplementedError – If the subclass does not implement sampling.

standardize(samples)[source]

Standardize samples using the prior mean and standard deviation.

Parameters:: samples (torch.Tensor) – Samples to standardize.
Returns:: Standardized samples.
Return type:: torch.Tensor

property std

Standard deviation of the prior.

Returns:: Standard deviation estimated from sampled values.
Return type:: torch.Tensor

class fast_lisa_subtraction.priors.analytical.Uniform(minimum, maximum, name=None, device='cpu')[source]

Bases: Prior

Uniform prior distribution.

Parameters:

minimum (float) – Minimum value of the prior.
maximum (float) – Maximum value of the prior.
name (str, optional) – Name of the prior parameter.
device (str, optional) – Torch device used for sampling.

sample(num_samples, standardize=False)[source]

Sample from the uniform prior.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior.

Return type:

torch.Tensor

class fast_lisa_subtraction.priors.analytical.UniformCosine(minimum=0, maximum=3.141592653589793, name=None, device='cpu')[source]

Bases: Prior

Angle prior uniform in the cosine of the angle.

This is commonly used for inclination angles.

Parameters:

minimum (float, optional) – Minimum angle in radians.
maximum (float, optional) – Maximum angle in radians.
name (str, optional) – Name of the prior parameter.
device (str, optional) – Torch device used for sampling.

Notes

Uniformity in \(\cos(\theta)\) implies a density proportional to \(\sin(\theta)\) over the support.

sample(num_samples, standardize=False)[source]

Sample from a cosine-uniform angle prior.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior.

Return type:

torch.Tensor

class fast_lisa_subtraction.priors.analytical.UniformSine(minimum=-1.5707963267948966, maximum=1.5707963267948966, name=None, device='cpu')[source]

Bases: Prior

Angle prior uniform in the sine of the angle.

This is commonly used for ecliptic latitude angles.

Parameters:

minimum (float, optional) – Minimum angle in radians.
maximum (float, optional) – Maximum angle in radians.
name (str, optional) – Name of the prior parameter.
device (str, optional) – Torch device used for sampling.

Notes

Uniformity in \(\sin(\theta)\) implies a density proportional to \(\cos(\theta)\) over the support.

sample(num_samples, standardize=False)[source]

Sample from a sine-uniform angle prior.

Parameters:

num_samples (int) – Number of samples to draw.
standardize (bool, optional) – Whether to standardize the samples to zero mean and unit variance (for training purposes). Default: False

Returns:

Samples drawn from the prior.

Return type:

torch.Tensor