draco.analysis.svdfilter

A set of tasks for SVD filtering the m-modes.

Functions

svd_em(A, mask[, niter, rank, full_matrices])

Perform an SVD with missing entries using Expectation-Maximisation.

Classes

`SVDFilter`()	SVD filter the m-modes to remove the most correlated components.
`SVDSpectrumEstimator`()	Calculate the SVD spectrum of a set of m-modes.

class draco.analysis.svdfilter.SVDFilter[source]

Bases: SingleTask

SVD filter the m-modes to remove the most correlated components.

niter

Number of iterations of EM to perform.

Type:: int

local_threshold

Cut out modes with singular value higher than local_threshold times the largest mode on each m.

Type:: float

global_threshold

Remove modes with singular value higher than global_threshold times the largest mode on any m

Type:: float

Initialize pipeline task.

May be overridden with no arguments. Will be called after any config.Property attributes are set and after ‘input’ and ‘requires’ keys are set up.

process(mmodes)[source]

Filter MModes using an SVD.

Parameters:: mmodes (container.MModes) – MModes to process
Returns:: mmodes
Return type:: container.MModes

class draco.analysis.svdfilter.SVDSpectrumEstimator[source]

Bases: SingleTask

Calculate the SVD spectrum of a set of m-modes.

niter

Number of iterations of EM to perform.

Type:: int

Initialize pipeline task.

May be overridden with no arguments. Will be called after any config.Property attributes are set and after ‘input’ and ‘requires’ keys are set up.

process(mmodes)[source]

Calculate the spectrum.

Parameters:: mmodes (containers.MModes) – MModes to find the spectrum of.
Returns:: spectrum
Return type:: containers.SVDSpectrum

draco.analysis.svdfilter.svd_em(A, mask, niter=5, rank=5, full_matrices=False)[source]

Perform an SVD with missing entries using Expectation-Maximisation.

This assumes that the matrix is well approximated by only a few modes in order fill the missing entries. This is probably not a proper EM scheme, but is not far off.

Parameters:

A (np.ndarray) – Matrix to SVD.
mask (np.ndarray) – Boolean array of masked values. Missing values are True.
niter (int, optional) – Number of iterations to perform.
rank (int, optional) – Set the rank of the approximation used to fill the missing values.
full_matrices (bool, optional) – Return the full span of eigenvectors and values (see scipy.linalg.svd for a fuller description).

Returns:

u, sig, vh – The singular values and vectors.

Return type:

np.ndarray