Navigation

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

class Engine( object )	Source

Engine defines common properties of all Engines.

An Engine moves around a walker in a random way such that its likelood remain above the low-likelihood-limit.

Attributes

• walkers : WalkerList
       list of walkers to be diffused

• errdis : ErrorDistribution
       error distribution to be used

• slow : int
       If slow > 0, run this engine every slow-th iteration.

• phantoms : PhantomCollection
       Collection of valid walker positions collected during engine execution

• maxtrials : int
       maximum number of trials for various operations

• rng : numpy.random.RandomState
       random number generator

• verbose : int
       if verbose > 4 report about the engines. (mostly for debugging)

• report : list of int (read only)
       reports number of succes, accepted, rejected, failed calls. Plus the total.

• unitRange : array_like (read only)
       present max size of the parameter cloud (in unitspace: [0,1])

• unitMin : array_like (read only)
       present minimum values of the parameter cloud (in unitspace: [0,1])

Author Do Kester.

Engine( walkers, errdis, slow=None, phantoms=None, copy=None, seed=4213, verbose=0 )

Constructor.

Only one PhantomCollection should be present for all Engines.

Parameters

• walkers : list of Walker
       walkers to be diffused
• errdis : ErrorDistribution
       error distribution to be used
• slow : None or int > 0
       Run this engine every slow-th iteration. None for all.
• phantoms : None or PhantomCollection
       Container for all valid walkers, that have been tried. But were not kept.
       To calculate the spread of the parameters vs likelihood.
• seed : int
       for random number generator
• verbose : int
       report about the engines when verbose > 4
• copy : Engine
       engine to be copied

copy( )

Return a copy of this engine.

bestBoost( problem, myFitter=None )

When a logL is found better that all the rest, try to update it using a fitter.

Parameters problem : Problem
     the problem at hand myFitter : None or Fitter
     None fetches LevenbergMarquardtFitter
     a (non-linear) fitter

setWalker( kw, problem, allpars, logL, walker=None, fitIndex=None )

Update the walker with problem, allpars, LogL and logW.

Parameters

• walker : Sample
       sample to be updated

• kw : int
       index in walkerlist, of the walker to be replaced

• problem : Problem
       the problem in the walker

• allpars : array_like
       list of all parameters

• logL : float
       log Likelihood

• walker : Walker or None
       Copy this walker or create new one

• fitIndex : array_like
       (new) fitIndex

noBoost( walker )

doBoost( walker )

Check if walker is best in phantoms and try to optimize.

Parameters

• walker : Walker
       new walker to be checked

domain2Unit( problem, dval, kpar=None )

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

Parameters

• problem : Problem
       the problem involved
• dval : float
       domain value for the selected parameter
• kpar : None or array_like
       selected parameter index, where kp is index in [parameters, hyperparams]
       None means all

unit2Domain( problem, uval, kpar=None )

Return domain value for the selected parameter.

Parameters

• problem : Problem
       the problem involved
• uval : array_like
       unit value for the selected parameter
• kpar : None or array_like
       selected parameter indices, where kp is index in [parameters, hyperparams]
       None means all.

startJourney( unitStart )

Calculate the starting position and reset

Parameters

• unitStart : array_like
       start position in npars-dimensions in unit space

calcJourney( unitDistance )

Calculate the distance travelled since reset

Parameters

• unitDistance : array_like
       step size in npars-dimensions in unit space

reportJourney( )

makeIndex( np, val )

reportCall( )

Store a call to engine

reportSuccess( )

Add 1 to the number of succesfull steps: logL < lowLhood.

reportReject( )

Add 1 to the number of rejected steps: logL > lowLhood.

reportFailed( )

Add 1 to the number of failed steps: could not construct a step.

reportBest( )

Add 1 to the number of best likelihoods found upto now.

printReport( best=False )

successRate( )

Return percentage of success.

getUnitMinmax( problem, lowLhood )

Calculate unit minimum and maximum from the Phantoms

Parameters

• problem : Problem
       To extract the unit range for
• lowLhood : float
       low likelihood boundary

getUnitRange( problem, lowLhood )

Calculate unit range and minimum from PhantomCollection

Parameters

• problem : Problem
       To extract the unit range for
• lowLhood : float
       low likelihood boundary

execute( kw, lowLhood )

Execute the engine for difusing the parameters

Parameters

• kw : walker-id
       walker to diffuse
• lowLhood : float
       low limit on the loglikelihood

Returns

• int : number of succesfull moves