pgmuvi.constraints

Kernel-specific constraint helpers for pgmuvi.

This module provides convenience functions for constructing gpytorch constraints that are physically motivated and based on properties of the observed data, such as sampling cadence and observation span.

pgmuvi.constraints.get_constraint_set(name)

Return the constraint-set dict for name.

Parameters:

name (str) – Name of the constraint set (e.g. "LPV").

Returns:

Mapping of parameter names to their bound specifications.

Return type:

dict

Raises:

ValueError – If name is not a recognised constraint set.

Examples

>>> from pgmuvi.constraints import get_constraint_set
>>> cs = get_constraint_set("LPV")
>>> cs["period"]["lower"]
(20.0, True)
>>> cs["period"]["upper"]
(None, False)
pgmuvi.constraints.lengthscale_constraint(span, min_fraction=0.01, max_fraction=2.0)

Constraint on a kernel lengthscale, expressed as fractions of a span.

Works for both temporal and wavelength lengthscales — pass the relevant axis range (e.g. data_span for time, wl_span for wavelength) and the bounds are computed as fractions of that span.

Parameters:

  • span (float) – Total range of the axis being constrained (e.g. max_time - min_time, or max_wavelength - min_wavelength).

  • min_fraction (float, optional) – Minimum lengthscale as a fraction of span, by default 0.01. Must be less than max_fraction.

  • max_fraction (float, optional) – Maximum lengthscale as a multiple of span, by default 2.0.

Returns:

Constraint suitable for registering on a lengthscale parameter.

Return type:

gpytorch.constraints.Interval

Raises:

ValueError – If min_fraction >= max_fraction, or if span <= 0.

Examples

Temporal lengthscale with a 100-day dataset:

>>> from pgmuvi.constraints import lengthscale_constraint
>>> c = lengthscale_constraint(span=100.0)
>>> c.lower_bound < c.upper_bound
True

Wavelength lengthscale with a 500 nm range:

>>> c = lengthscale_constraint(span=500.0, min_fraction=0.01, max_fraction=10.0)
>>> c.lower_bound < c.upper_bound
True
pgmuvi.constraints.outputscale_constraint(data_std, min_factor=0.001, max_factor=100.0)

Constraint on an output scale (amplitude) parameter.

Parameters:

  • data_std (float) – Standard deviation of the observed data.

  • min_factor (float, optional) – Minimum output scale as a multiple of data_std, by default 0.001.

  • max_factor (float, optional) – Maximum output scale as a multiple of data_std, by default 100.0.

Return type:

gpytorch.constraints.Interval

Examples

>>> from pgmuvi.constraints import outputscale_constraint
>>> c = outputscale_constraint(data_std=1.0)
>>> c.lower_bound < c.upper_bound
True
pgmuvi.constraints.period_constraint(data_span, min_period_fraction=0.05, max_factor=1.0)

Constraint on a period parameter based on the data time span.

Restricts the period to be at least min_period_fraction * data_span and at most max_factor * data_span.

The default minimum-period fraction of 0.05 (5 % of the data span) is physically motivated by the typical observational requirements for cool evolved stars — the primary target class for pgmuvi — whose periods are generally tens of days or longer. For a 500-day dataset this implies a lower bound of ~25 days; for a 1000-day dataset ~50 days. Adjust min_period_fraction for other target classes or shorter baselines.

Parameters:

  • data_span (float) – Total duration of the observations (max time - min time), in whatever units the input time axis uses (days, normalised, etc.).

  • min_period_fraction (float, optional) – Minimum period as a fraction of data_span, by default 0.05.

  • max_factor (float, optional) – Maximum period as a multiple of data_span, by default 1.0.

Returns:

Constraint suitable for registering on a period parameter.

Return type:

gpytorch.constraints.Interval

Examples

>>> from pgmuvi.constraints import period_constraint
>>> c = period_constraint(data_span=100.0)
>>> c.lower_bound < c.upper_bound
True
pgmuvi.constraints.positive_constraint()

Return a simple positivity constraint.

Return type:

gpytorch.constraints.Positive

pgmuvi.constraints.wavelength_constraint(wl_span, min_fraction=0.01)

Constraint on a wavelength kernel lengthscale.

Alias for lengthscale_constraint() with max_fraction=10.0, kept for backward compatibility.

Parameters:

  • wl_span (float) – Total wavelength range (max wavelength - min wavelength).

  • min_fraction (float, optional) – Minimum lengthscale as a fraction of the wavelength span, by default 0.01.

Returns:

Constraint suitable for a wavelength kernel lengthscale parameter.

Return type:

gpytorch.constraints.Interval

Examples

>>> from pgmuvi.constraints import wavelength_constraint
>>> c = wavelength_constraint(wl_span=500.0)
>>> c.lower_bound < c.upper_bound
True

See also

lengthscale_constraint

General-purpose lengthscale constraint.