InformationBudget

class ell1fit.information.InformationBudget(parameter_names: list, segments: list)

Bases: object

Every segment’s contribution to one set of shared parameters.

Attributes:

parameter_nameslist of str

The shared parameters, in the order they index every matrix here.

segmentslist of SegmentInformation

One entry per observation.

Attributes Summary

parameter_names
segments
total	Summed effective information: the precision matrix of the shared parameters.

Methods Summary

correlation([fixed])	Correlation matrix of the free shared parameters.
covariance([fixed])	Covariance of the shared parameters, optionally holding some of them fixed.
cumulative(parameter)	Names in ranked order, and the running sum of their information shares.
n_segments_for(parameter, fraction)	How many of the best segments carry this fraction of the information.
ranking(parameter)	Each segment's share of the diagonal information on one parameter.
uncertainties([fixed])	Predicted 1-sigma uncertainty per shared parameter, as a dictionary.
without(name)	A budget with one segment removed, for leave-one-out questions.

Attributes Documentation

parameter_names: list = <dataclasses._MISSING_TYPE object>

segments: list = <dataclasses._MISSING_TYPE object>

total

Summed effective information: the precision matrix of the shared parameters.

Methods Documentation

correlation(fixed=())

Correlation matrix of the free shared parameters.

covariance(fixed=())

Covariance of the shared parameters, optionally holding some of them fixed.

Holding a parameter fixed removes its row and column before the inversion, which is the difference between “we did not fit it” and “we fitted it and then ignored it”.

cumulative(parameter)

Names in ranked order, and the running sum of their information shares.

n_segments_for(parameter, fraction)

How many of the best segments carry this fraction of the information.

ranking(parameter)

Each segment’s share of the diagonal information on one parameter.

Returns (name, fraction) pairs, largest first, summing to one. This is a budget, not a sensitivity: for what actually happens to the error bar when a segment goes away, use without().

uncertainties(fixed=())

Predicted 1-sigma uncertainty per shared parameter, as a dictionary.

without(name)

A budget with one segment removed, for leave-one-out questions.