[vtkusers] K-means values

[David Thompson]

Hi Sara,

> Update: from the vtkKMeansStatistics reference page it sounds like  
> it's necessary for the first column to be a vtkIdType array in order  
> for learning to be used and the table available on the output port.

Only the first column of a table on the LEARN_PARAMETERS port needs to  
be vtkIdTypeArray. In the code below, the inputData table is attached  
to the INPUT_DATA port, so there's no constraint on the column type.  
I'm not sure I understand exactly what the code is supposed to  
accomplish; the column you create named "KMeans" is initially going to  
be filled with zeros. The other column will be named  
FeatureInArrayName (not "ScalarMagnitude" because your don't ever call  
inputData->AddColumn on floatArray). Since all the values in the first  
column are identical and both Learn and Assess are turned on, the  
filter will
1. Perform Learn by randomly choosing "clusterCount" cluster centers  
(since no LEARN_PARAMETERS input was set) and computing a single  
iteration of k-means.
2. Perform Assess by adding 2 columns to the input table: one  
vtkDoubleArray containing the squared distance to the nearest cluster  
center, and one vtkIdTypeArray containing the nearest cluster center's  
ID (as reported by the OUTPUT_MODEL output).
The output model kMeansStatistics->GetOutputDataObject(1)->GetBlock(0)  
will report all of the cluster centers as lying on the line "KMeans"=0.
Is that what you had in mind?

	David

> 	//get data from the file
> 	vtkXMLPolyDataReader* reader = vtkXMLPolyDataReader::New();
> 	reader->SetFileName( inputFile );
> 	reader->Update();
> 	vtkPolyData* polydata = reader->GetOutput();
> 	cout << "Reading data" << endl;
> 	
> 	vtkSmartPointer<vtkFloatArray> magnitudeArray =  
> vtkFloatArray::SafeDownCast(polydata->GetPointData()- 
> >GetArray(FeatureInArrayName));
> 	
> 	// set up table for KMeans
> 	vtkSmartPointer<vtkTable> inputData =  
> vtkSmartPointer<vtkTable>::New();
> 	
> 	// first column for kmeans clusters, vtkIdType
> 	vtkSmartPointer<vtkIdTypeArray> idArray =  
> vtkSmartPointer<vtkIdTypeArray>::New();
>     	idArray->SetNumberOfComponents(1);
>    	idArray->SetName( "KMeans" );
>     	idArray->SetNumberOfTuples(polydata->GetNumberOfPoints());
> 	inputData->AddColumn( idArray );
> 	
> 	// second column for data values
> 	vtkSmartPointer<vtkFloatArray> floatArray =  
> vtkSmartPointer<vtkFloatArray>::New();
> 	floatArray->SetNumberOfComponents(1);
> 	floatArray->SetName( "ScalerMagnitude" );
> 	floatArray->SetNumberOfTuples(polydata->GetNumberOfPoints());
> 	inputData->AddColumn( magnitudeArray );
> 	
> 	// KMeans Set up
> 	vtkSmartPointer<vtkKMeansStatistics> kMeansStatistics =  
> vtkSmartPointer<vtkKMeansStatistics>::New();
> 	
> 	kMeansStatistics->SetInput( vtkStatisticsAlgorithm::INPUT_DATA,  
> inputData );
> 	kMeansStatistics->SetColumnStatus( inputData->GetColumnName( 0 ) ,  
> 1 );
> 	kMeansStatistics->SetColumnStatus( inputData->GetColumnName( 1 ) ,  
> 1 );
> 	kMeansStatistics->RequestSelectedColumns();
> 	kMeansStatistics->SetDefaultNumberOfClusters( clusterCount );
> 	kMeansStatistics->SetAssessOption( 1 );	
> 	kMeansStatistics->SetLearnOption( 1 );
> 	kMeansStatistics->SetMaxNumIterations( 1 );
> 	kMeansStatistics->Update() ;
> 	
> 	//vtkTable* tab = vtkTable::SafeDownCast( kMeansStatistics- 
> >GetOutputDataObject( 1 ).GetBlock( 0 ) );
> 	//double xc = tab->GetValueByName( label, "x" ).ToDouble();
> 	kMeansStatistics->GetOutputDataObject(1).GetBlock( 0 );
>
>
> On Mar 18, 2011, at 4:30 PM, Sara Rolfe wrote:
>
>> I agree.  I also think that the example on the  wiki might be more  
>> helpful if it showed this method of accessing the means.
>>
>> However, I'm still having some trouble accessing the first block of  
>> the k-means output, when I use:
>>
>> 	vtkTable* tab = vtkTable::SafeDownCast( kMeansStatistics- 
>> >GetOutputDataObject( 1 ).GetBlock( 0 ) );
>>
>> or just
>> 	kMeansStatistics->GetOutputDataObject( 1 ).GetBlock( 0 )
>>
>> I'm getting the following error:
>>
>> error: request for member ‘GetBlock’ in ‘kMeansStatistics.  
>> vtkSmartPointer<T>::operator-> [with T = vtkKMeansStatistics]()- 
>> > 
>> vtkKMeansStatistics 
>> ::< 
>> anonymous 
>> > 
>> .vtkStatisticsAlgorithm 
>> ::< 
>> anonymous 
>> > 
>> .vtkTableAlgorithm 
>> ::<anonymous>.vtkAlgorithm::GetOutputDataObject(1)’, which is of  
>> non-class type ‘vtkDataObject*’
>>
>> Does anyone know what might cause this?
>>
>> Thanks,
>> Sara
>>
>> On Mar 16, 2011, at 5:32 PM, David Doria wrote:
>>
>>> On Wed, Mar 16, 2011 at 4:07 PM, Sara Rolfe <smrolfe at u.washington.edu 
>>> > wrote:
>>>>
>>>> Hi David,
>>>>
>>>> It's not a difficult fix, I mainly found it awkward since I  
>>>> thought I was missing something simple.  I understand how it  
>>>> works better now and can certainly implement it this way.  I  
>>>> appreciate your clarification.
>>>>
>>>> Sara
>>>
>>> If you recall, I had the same sort of "it feels awkward" issues a
>>> while back. I think the problem is that the suite is very very  
>>> capable
>>> and able to hand very large and serious problems, but many users  
>>> just
>>> want the "very basic" functionality ("which points belong to which
>>> cluster?", and "what are the cluster means?"), which is hard to get
>>> at. There are many layers of complexity above that that a user must
>>> understand (i.e. this "Learn" and "Derive" that you mentioned) that
>>> are not concepts in a standard textbook explanation of kmeans
>>> clustering. Providing a "simple" interface to do this "one pass"  
>>> type
>>> of thing would definitely be a nice addition.
>>>
>>> That is, replace:
>>>
>>>  kMeansStatistics->SetLearnOption( 1 ); // This is on by default.
>>>  kMeansStatistics->SetMaxNumIterations( 1 );
>>>  vtkTable* tab = vtkTable::SafeDownCast(
>>> kMeansStatistics->GetOutputDataObject( 1 ).GetBlock( 0 ) );
>>>  double xc = tab->GetValueByName( label, "x" ).ToDouble();
>>>
>>> with
>>>
>>> double mean0[3];
>>> kMeans->GetMean(0,mean0);
>>>
>>> See the difference :) ?
>>>
>>> David D.
>>
>> _______________________________________________
>> Powered by www.kitware.com
>>
>> Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html
>>
>> Please keep messages on-topic and check the VTK FAQ at: http://www.vtk.org/Wiki/VTK_FAQ
>>
>> Follow this link to subscribe/unsubscribe:
>> http://www.vtk.org/mailman/listinfo/vtkusers
>