Difference between revisions of "Known bugs"

From SUMOwiki
Jump to navigationJump to search
 
(34 intermediate revisions by 4 users not shown)
Line 1: Line 1:
 
While we always try to test each release as much as we can, inevitably some bugs will always slip through unnoticed.  This page only shows the most important bugs that have surfaced after the release was made.
 
While we always try to test each release as much as we can, inevitably some bugs will always slip through unnoticed.  This page only shows the most important bugs that have surfaced after the release was made.
  
Remember to also check the bugs on the [http://www.sumo.intec.ugent.be project website].  To stay up to date with the latest news and releases, we also recommend subscribing to our newsletter [http://www.sumo.intec.ugent.be here].  Traffic will be kept to a minimum and you can unsubscribe at any time.   
+
To stay up to date with the latest news and releases, we also recommend subscribing to our newsletter [http://www.sumo.intec.ugent.be here].  Traffic will be kept to a minimum and you can unsubscribe at any time.   
  
 
If you happen to encounter something not listed here, please [[reporting problems|report it]].
 
If you happen to encounter something not listed here, please [[reporting problems|report it]].
 +
 +
== Version 6.2 ==
  
  
== Version 5.0 ==
 
 
{| border="0" style="text-align:left"
 
{| border="0" style="text-align:left"
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| Duplicate samples and samples with NaN/Inf values are not filtered properly.
+
| There is a small error in LRMMeasure that slows the modeling process if LRM is used together with another measure.
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
| Fixed in SVN.
+
| Fixed in next version
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| Wait for next release. Simulators that may return invalid values or duplicate samples should be avoided until then.
+
| In src/matlab/measures/@LRMMeasure/calculateMeasure.m, replace line 204 by: <source lang="matlab">dtot(:,k) = (sum(d,1) ./ numTp) + dtot(:,k);</source>
 
|-
 
|-
 
|}
 
|}
  
  
== Version 4.2 ==
+
== Version 6.1.1 ==
 +
 
 +
 
 
{| border="0" style="text-align:left"
 
{| border="0" style="text-align:left"
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| Loading the random state from file gives a directory does not exist error.
+
| When modeling complex data, depending on the measure/error function used, the model generation process my be sub-optimal in some cases
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
| Fixed in SVN.
+
| Fixed in next version
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| Upgrade to the next version.
+
| Execute the error function you use (default = beeq) with a couple of complex numbers.  The result must be a real number (non-complex).  If this is not the case, fix the function (or ask us to send you a new one) or use a different function (also make sure to test it).
 
|-
 
|-
 
|}
 
|}
  
 +
== Version 6.1 ==
  
 
{| border="0" style="text-align:left"
 
{| border="0" style="text-align:left"
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| When starting the toolbox for the first time you get an error like "The class ContextConfig cannot be found, make sure java is running"
+
| When using the LS-SVM models and you have not compiled the mex files the toolbox crashes with "Too many open files".
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
| Fixed in SVN.
+
| Fixed in next version
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| This is a bug in Matlab itself. Please [[contact]] us for instructions on how to fix this.
+
| In /src/matlab/contrib/LS-SVMlab/trainlssvm.m replace on line 259 the text <source lang="bash">'model.implementation=''CFILE''</source> with the text <source lang="bash">'model.implementation=''MATLAB''</source>
 
|-
 
|-
 
|}
 
|}
 +
  
  
Line 53: Line 58:
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| When starting the toolbox for the first time you get an error like "The class be.ac.ua.coms.m3.config.Util has no field or method named genHash"
+
| The ParetoModelBuilder can crash with "unknown function createModelFromIndividual()".
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
| Fixed in SVN.
+
| Fixed in next version
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| This is a bug in Matlab itself. Please [[contact]] us for instructions on how to fix this.
+
| Replace the offending line (line 70 in modelbuilders/@ParetoModelbuilder/runLoop.m) with: "model = createModel(getModelFactory(s), populationEntry);".  Make also sure that you have set the option ''paretoMode="true"''.
 
|-
 
|-
 
|}
 
|}
Line 67: Line 72:
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| When the TestSamples measure is used without a dataset the testset is not calculated correctly!!
+
| When building models multi-objectively the toolbox uses a lot of diskspace and signals a lot of new best models
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
| Fixed in SVN.
+
| Fixed in next version
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| Replace the line 'tmp(testSet,:) = [];' by 'tmp(testSet,:) = NaN;'
+
| Remove in modelbuilders/@AdaptiveModelbuilder/private/orderBestModels.m, the condition "length(s.bestModels) == 1" so the if looks like "if ( oldBestModelId ~= newBestModelId )"
 
|-
 
|-
 
|}
 
|}
  
 +
 +
<!--
 +
 +
== Version 6.0.1 ==
  
 
{| border="0" style="text-align:left"
 
{| border="0" style="text-align:left"
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| The optimization framework may fail in specific circumstances
+
| getExpression does not work for rational models with complex outputs. It produces output different from evaluate, and evaluate is the one producing the correct output.
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
| The framework is currently under redesign and a much improved, more robust, version will be released with 5.0
+
| Fixed in next version
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| Upgrade to 5.0 when it becomes available.
+
| Please wait for release
 
|-
 
|-
 
|}
 
|}
  
 +
== Version 6.0 ==
  
 
{| border="0" style="text-align:left"
 
{| border="0" style="text-align:left"
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| The BatchModelBuilder fails with an error
+
| You get an error like "No appropriate method or public field isProjectMode for class ibbt.sumo.config.ContextConfig"
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
| Fixed in SVN
+
| Fixed in 6.0.1
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| Simply remove the offending line
+
| Please upgrade to 6.0.1
 
|-
 
|-
 
|}
 
|}
  
== Version 4.1 ==
+
 
 
{| border="0" style="text-align:left"
 
{| border="0" style="text-align:left"
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| The maximum number of samples is not respected in a correct way
+
| The Kriging models do not work with combineOutputs=true in some cases
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
| Fixed in SVN.
+
| Fixed in SVN
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| Upgrade to 4.2 when it becomes available.
+
| Please upgrade to 6.0.1
 
|-
 
|-
 
|}
 
|}
Line 123: Line 133:
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| The toolbox crashes when you pass samples on the command line
+
| The automatic model type selection algorithm (heterogeneous evolution) does not work with Matlab 2008a and later
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
| Fixed in SVN.
+
| Fixed in SVN
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| On the offending line replace "scale(..)" by "scaleColumns(...)".
+
| Use an older Matlab version (e.g., 2007a)
 
|-
 
|-
 
|}
 
|}
  
== Version 4.0 ==
+
== Version 5.0 ==
{| border="0" style="text-align:left"
 
|-
 
!'''Bug'''
 
| The toolbox goes into a seemingly infinite loop if u are using the Matlab sample evaluator and there is an error in your matlab function.
 
|-
 
!'''Status'''
 
| Wont fix
 
|-
 
!'''Workaround'''
 
| This is due to the poor Java -> Matlab interface and thus there is little we can do about it. It is best to make sure your function runs without error by running it manually.
 
|-
 
|}
 
  
 +
We have found some important bugs in 5.0 that affect various parts of the model generation process. This could mean (in some cases) that the final models you get are not really as good as they could be.  This has been fixed in 6.0 and we are working hard to release it as soon as possible.
  
 
{| border="0" style="text-align:left"
 
{| border="0" style="text-align:left"
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| In some rare cases the toolbox goes into an infinite loop when using the grid sample selector.
+
| Consecutive runs on the same dataset are not independent (can cause dataset depleted exceptions)
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
Line 157: Line 156:
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| This happens in some rare cases, and is due to the way the grid SE works.  This will be fixed in the next release.  In the meantime, if you run into the problem use a different sample selector or combine the grid SE with a different SE (e.g., the random SE) using the CombinedSampleSelector.
+
| Wait for next release.
 
|-
 
|-
 
|}
 
|}
Line 165: Line 164:
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| If the first component inside the <Plan> tag is a component that is overridden in the <Run> tag, a NullPointerException occurs
+
| Duplicate samples and samples with NaN/Inf values are not filtered properly (for example you get Matrix dimension mismatch errors).
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
Line 171: Line 170:
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| This is a very obscure bug, simply make sure that the first component in your <Plan> tag is one that is NOT overridden in a <Run> tag.  Use for example <ContextConfig>.
+
| Wait for next release or simply disable the relevant code. Simulators that may return invalid values or duplicate samples should be avoided until then.
 
|-
 
|-
 
|}
 
|}
Line 179: Line 178:
 
|-
 
|-
 
!'''Bug'''
 
!'''Bug'''
| When modeling the same output multiple times in parallel, some profilers may interfere with each other
+
| You get the following error: "Failed to create object of type ANNGeneticInterface, error is "Undefined variable "logger" or class "logger.severe"
 
|-
 
|-
 
!'''Status'''  
 
!'''Status'''  
Line 185: Line 184:
 
|-
 
|-
 
!'''Workaround'''  
 
!'''Workaround'''  
| Avoid modeling the same output multiple times in the same run.  Try using different runs or turning off the profilers that give problems.
+
| Simply remove the offending line or set complex handling to split or modulus (you are trying to model complex data with ANNs).
|-
 
|}
 
 
 
== Version 3.3 ==
 
{| border="0" style="text-align:left"
 
|-
 
!'''Bug'''
 
| Loading a saved model file from disk gives warnings and in some cases even errors (see the [[FAQ]])
 
|-
 
!'''Status'''
 
| fixed in the latest snapshots
 
|-
 
!'''Workaround'''
 
| if you are lucky you can just ignore the warnings and it should work, in some cases it wont and you will need to upgrade
 
 
|-
 
|-
 
|}
 
|}
  
 
+
-->
{| border="0" style="text-align:left"
 
|-
 
!'''Bug'''
 
| Movie generation gives problems on windows
 
|-
 
!'''Status'''
 
| Fixed
 
|-
 
!'''Workaround'''
 
| Use a different operating system (Linux, OSX, ...) or upgrade and edit ''src/matlab/modelbuilders/@AdaptiveModelBuilder/createMovie.m'' file so it creates an avi instead of a mov file (Warning: the avi file is not compressed and may be very big!)
 
|-
 
|}
 

Latest revision as of 13:25, 8 October 2009

While we always try to test each release as much as we can, inevitably some bugs will always slip through unnoticed. This page only shows the most important bugs that have surfaced after the release was made.

To stay up to date with the latest news and releases, we also recommend subscribing to our newsletter here. Traffic will be kept to a minimum and you can unsubscribe at any time.

If you happen to encounter something not listed here, please report it.

Version 6.2

Bug There is a small error in LRMMeasure that slows the modeling process if LRM is used together with another measure.
Status Fixed in next version
Workaround In src/matlab/measures/@LRMMeasure/calculateMeasure.m, replace line 204 by:
dtot(:,k) = (sum(d,1) ./ numTp) + dtot(:,k);


Version 6.1.1

Bug When modeling complex data, depending on the measure/error function used, the model generation process my be sub-optimal in some cases
Status Fixed in next version
Workaround Execute the error function you use (default = beeq) with a couple of complex numbers. The result must be a real number (non-complex). If this is not the case, fix the function (or ask us to send you a new one) or use a different function (also make sure to test it).

Version 6.1

Bug When using the LS-SVM models and you have not compiled the mex files the toolbox crashes with "Too many open files".
Status Fixed in next version
Workaround In /src/matlab/contrib/LS-SVMlab/trainlssvm.m replace on line 259 the text
'model.implementation=''CFILE''
with the text
'model.implementation=''MATLAB''


Bug The ParetoModelBuilder can crash with "unknown function createModelFromIndividual()".
Status Fixed in next version
Workaround Replace the offending line (line 70 in modelbuilders/@ParetoModelbuilder/runLoop.m) with: "model = createModel(getModelFactory(s), populationEntry);". Make also sure that you have set the option paretoMode="true".


Bug When building models multi-objectively the toolbox uses a lot of diskspace and signals a lot of new best models
Status Fixed in next version
Workaround Remove in modelbuilders/@AdaptiveModelbuilder/private/orderBestModels.m, the condition "length(s.bestModels) == 1" so the if looks like "if ( oldBestModelId ~= newBestModelId )"