[vtkusers] PCA visualizer app

[Rahul Chander]

Hi,
I am currently building a PCA visualizer application using VTK. This 
application has inputs the eigenvectors of a PCA analysis.  With the 
help of slider(s) in the GUI, a user can manipulate the coeff. for 
the basis vectors and the resulting superposition of all components 
is visualized on the screen.   
Now the issue:
Currently, I let the user input the coeff thru a slider in the GUI 
and then I do this calculation:

[x,y,z]=[coeff1*eig1+coeff2*eig2+coeff3*eig3 .....and so on]
Then, I add the [x,y,z] from above into a vtkPointSource variable 
using the InsertNextPoint() method.  
I have to repeat this everytime a coeff value changes, so the points 
are re-entered and then rendered using the surface generation for a 
datacloud.  The result is slow.  It takes too long for the slider 
interaction.  

I think the situation can be improved by using pipeline mechanism and 
using the methods VTK provides to do this calculation.
[x,y,z]=[coeff1*eig1+coeff2*eig2+coeff3*eig3 .....and so on]

The problem is, I do not know which method in VTK supports a vector 
addition so that I dont have to reinitialize using the PointSource 
mechanism.  Most methods are supported for only 3 dim unlike ITK 
which is supported for N-Dim.  Even though each basis vector has 
only 3 components, the #basis components more than 3.  

I was wondering if a Transform method could be used for the matrix 
multiplication part, but again the issue of N-Dims .  The number of 
points remain the same, only their position gets recalculated.

Here is some portion of the Python code:
eig_x is the x component of all the eigenvectors
eig_y is the y component of all .....
eig_z is the z ....
pointSource is the set of points which gets rendered. type is 
vtkPolyData.  It is essentially a datacloud and a surface is 
generated.
Dot() is a function I wrote that takes the dot-product of two vectors 
over N-dim

def readPoints():
	global coeff_t,eig_x,eig_y,eig_z,pointSource
	output = pointSource.GetPolyDataOutput()
	points = vtk.vtkPoints()
	output.SetPoints(points)
	for i in range(0, LENGTH):
		x = Dot(coeff_t[0],eig_x[i])
		y = Dot(coeff_t[1],eig_y[i])
		z = Dot(coeff_t[2],eig_z[i])
		points.InsertNextPoint(x,y,z)
	
Does anyone have any suggestions to improve the above function?
Thanks
Rahul
-- 
Rahul Chander, BASc (Electrical Eng.)

War is much too serious a matter to be entrusted to the military.
          -- Clemenceau