Navigation

Global	Documentation
Home	Manual
Readme	Glossary
Notes	Design
Examples	Classes
	Trouble
	Style

class ErrorDistribution( object )	Source

ErrorDistribution defines general methods for a error distribution.

Error distributions are used to calculate the likelihoods.

Author Do Kester.

Attributes

• hyperpar : HyperParameter
       hyperparameter for the error distribution
• deltaP : float
       delta for calculating numerical derivatives
• ncalls : int
       number of calls to the logLikelihood
• nparts : int
       number of calls to the partial of the logLikelihood
• sumweight : float
       sum over the weights or ndata
• ndata : int
       number of points in data
• hypar : [float]
       list of values for the hyperparameters
• nphypar : int
       number of hyper parameters in this error distribution
• constrain : None or callable
       None: Use logLikelihood as is
       callable: logL = func( logL, problem, allpars, lowLhood )
                 returning a (modified) value of the logLikelihood.

ErrorDistribution( fixed=None, constrain=None, copy=None )

Constructor.

Parameters

• fixed : dictionary of {int:float}
       int list if parameters to fix permanently. Default None.
       float list of values for the fixed parameters.
• constrain : None or callable
       function as: func( logL, problem, allpars )
       returning a (modified) value of the logLikelihood.
• copy : ErrorDistribution
       distribution to be copied.

Raise

ValueError when constrain is not a callable method.

copy( )

Return copy of this.

getGaussianScale( problem, allpars=None )

Return the noise scale.

*** Gaussian approximation ***

Parameters

• problem : Problem
       to be solved
• allpars : array_like
       None take parameters from problem.model
       list of all parameters in the problem

getResiduals( problem, allpars=None )

Return residuals: ydata - model.result

Parameters

• problem : Problem
       to be solved
• allpars : array_like
       None take parameters from problem.model
       list of all parameters in the problem

getChisq( problem, allpars=None )

Return chisq

*** Gaussian approximation ***

Sum over the (weighted) squared residuals

Parameters

• problem : Problem
       to be solved
• allpars : array_like
       None take parameters from problem.model
       list of all parameters in the problem

toSigma( scale )

Return sigma, the squareroot of the variance.

Parameter

• scale : float
       the scale of this distribution.

Return default value : scale

isBound( )

True when all priors of its (hyper)parameters are bound

acceptWeight( )

True if the distribution accepts weights.

keepFixed( fixed=None )

Keeps (hyper)parameters fixed at the provided values.

1. Repeated calls start from scratch.
   
2. Reset with keepFixed( fixed=None )

Parameters

• fixed : dictionary of {int:float}
       int list if parameters to fix permanently. Default None.
       float list of values for the fixed parameters.

setPriors( priors )

Set priors on the hyper parameter(s).

Parameters

• priors : (list of) Prior
       prior distribution for the hyperparameters

setLimits( limits )

Set limits on the hyper parameter(s).

Parameters

• limits : [low,high]
       low : float or array_like
           low limits
       high : float or array_like
           high limits

domain2Unit( dval, ks )

Return value in [0,1] for the selected parameter.

Parameters

• dval : float
       hyper parameter value in domain
• ks : int
       selecting index

unit2Domain( uval, ks )

Return domain value for the selected parameter.

Parameters

• uval : float
       unit value of hyper parameter
• ks : int
       selecting index

logCLhood( problem, allpars )

Return the constrained log( likelihood ).

Parameters

• problem : Problem
       to be solved
• allpars : array_like
       parameters of the problem

logLhood( problem, allpars )

Return the log( likelihood ).

Parameters

• problem : Problem
       to be solved
• allpars : array_like
       parameters of the problem

partialLogL( problem, allpars, fitIndex )

Return the partial derivative of log( likelihood ) to the parameters.

Parameters

• problem : Problem
       to be solved
• allpars : array_like
       parameters of the problem
• fitIndex : array_like
       indices of parameters to be fitted

partialLogL_alt( problem, allpars, fitIndex )

Return the partial derivative of log( likelihood ) to the parameters.

Alternative calculation.

Parameters

• problem : Problem
       to be solved
• allpars : array_like
       parameters of the problem
• fitIndex : array_like
       indices of parameters to be fitted

numPartialLogL( problem, allpars, fitIndex )

Return d log( likelihood ) / dp, numerically calculated.

Parameters

• problem : Problem
       to be solved
• allpars : array_like
       parameters of the problem
• fitIndex : array_like
       indices of parameters to be fitted

updateLogL( problem, allpars, parval=None )

Return a update of the log( likelihood ) given a change in a few parameter.

This method provides the opportunity to optimize the logL calculation. Providing this one, automatically provides the previous one. For now it just refers to logLikelihood() itself.

Parameters

• problem : Problem
       to be solved
• param : array_like
       parameters of the model
• parval : dict of {int : float}
       int index of a parameter
       float (old) value of the parameter

setResult( )

hyparname( k )

Return name of the hyperparameter

Parameters

• k : int
       index of the hyperparameter