MantisBT - ITK
View Issue Details
IDProjectCategoryView StatusDate SubmittedLast Update
0003606ITKpublic2006-08-14 08:452010-11-06 23:43
ReporterLuis Ibanez 
Assigned Tobrian avants 
PriorityurgentSeverityminorReproducibilityalways
StatusassignedResolutionopen 
PlatformOSOS Version
Product Version 
Target VersionFixed in Version 
Resolution Date
Sprint
Sprint Status
Summary0003606: RegularStepGradientDescentOptimizer reports current position out of sync in iteration event
DescriptionIn the RegularStepGradientDescentBaseOptimizer the IterationEvent is invoked at a moment where the current position has already been computed for the next iteration. This results in mismatch between the "current position" and the "m_Value" reported.

Lines 263 and 267 should be reverted.
Steps To Reproduce
Additional Information
TagsNo tags attached.
Relationships
related to 0003205assigned Ziv Yaniv Record the best value visited by an optimizer 
Attached Files
Issue History
Date ModifiedUsernameFieldChange
2010-11-05 00:26Cory W QuammenRelationship addedrelated to 0003205
2010-11-06 23:39Luis IbanezAssigned ToLuis Ibanez => brian avants
2010-11-06 23:43Luis IbanezNote Added: 0022935

Notes
(0004638)
Luis Ibanez   
2006-08-14 08:46   
We should check whether a similar problem happens in other optimizers.
(0004639)
Luis Ibanez   
2006-08-14 08:48   
There is also an issue with not keeping the position of the actual lowest value:

http://public.kitware.com/pipermail/insight-users/2006-August/019041.html [^]
(0022935)
Luis Ibanez   
2010-11-06 23:43   
This still seems to be the case.

itkRegularStepGradientDescentBaseOptimizer.cxx: 266

  this->InvokeEvent( IterationEvent() );

at the end of the AdvanceOneStep() method.

which means that the Current position is the new one,
but the cached value is the one of the previous position.