observation.realizations#

Concrete parameter values for observation models.

Classes

`ObsParamsRealization`	A single realization of observation model parameters.
`ObsParamsPoint`	Point estimates of observation model parameters.

Functions

adjust_snr

Scale observation precisions to achieve target signal-to-noise ratios.

class ObsParamsRealization( C: ndarray, d: ndarray, phi: ndarray, alpha: ndarray, y_dims: ndarray, x_dim: int, )[source]#

A single realization of observation model parameters.

Sources: prior sampling, posterior means, posterior samples.

Parameters:

Cndarray of float, shape (y_dim, x_dim): Loading matrices.
dndarray of float, shape (y_dim,): Observation means.
phindarray of float, shape (y_dim,): Observation precisions.
alphandarray of float, shape (n_groups, x_dim): ARD parameters.
y_dimsndarray of int, shape (n_groups,): Dimensionalities of each observed group.
x_dimint: Number of latent dimensions.

class ObsParamsPoint( C: ndarray, d: ndarray, phi: ndarray, y_dims: ndarray, x_dim: int, )[source]#

Point estimates of observation model parameters.

Source: Non-Bayesian fitting (FA, GPFA, etc.)

Semantically distinct from ObsParamsRealization—represents “the” optimized answer, not “a” sample from a distribution. Does not include alpha (non-Bayesian methods do not use ARD).

Parameters:

Cndarray of float, shape (y_dim, x_dim): Loading matrices.
dndarray of float, shape (y_dim,): Observation means.
phindarray of float, shape (y_dim,): Observation precisions.
y_dimsndarray of int, shape (n_groups,): Dimensionalities of each observed group.
x_dimint: Number of latent dimensions.

adjust_snr( realization: ObsParamsRealization, snr: float | ndarray, y_dims: ndarray | None = None, ) → ObsParamsRealization[source]#

Scale observation precisions to achieve target signal-to-noise ratios.

SNR is defined as var(signal) / var(noise), where signal variance comes from the loading matrices C and noise variance from observation precisions phi. This function scales phi to achieve the target SNR per group.

Parameters:

realizationObsParamsRealization: Observation parameters to adjust.
snrfloat or ndarray: Target SNR. Either a scalar (broadcast to all groups) or per-group array of shape (n_groups,).
y_dimsndarray of int, shape (n_groups,) or None, default None: Dimensionalities of each group. If None, uses realization.y_dims.

Returns:

ObsParamsRealization: New realization with adjusted phi values. Other parameters unchanged.