# Select columns of structured data
**URL:** https://discourse.paraview.org/t/select-columns-of-structured-data/8822
**Category:** ParaView Support
**Created:** [January 21, 2022, 7:38pm UTC](https://discourse.paraview.org/t/select-columns-of-structured-data/8822 "2022-01-21T19:38:50Z")
**Posts on this page:** 7
**Page:** 1
### Author:  [@woodscn](https://discourse.paraview.org/u/woodscn)
#### Post date: [January 21, 2022, 7:38pm UTC](https://discourse.paraview.org/t/select-columns-of-structured-data/8822/1 "2022-01-21T19:38:51Z")
I’m working with image data in a programmable filter, and I would like to work with entire columns or sheets of celldata (e.g. data[inda, indb, :], data[inda, :, :]) Is there a way for me to do this?
---
### Author:  [@pavel](https://discourse.paraview.org/u/pavel)
#### Post date: [January 25, 2022, 6:26am UTC](https://discourse.paraview.org/t/select-columns-of-structured-data/8822/2 "2022-01-25T06:26:17Z")
Hi Nathan,
Could you please clarify your question or provide some data? If you want to use all the values of one cell property, you can get them into numpy array. If you want to use all the values of all the cell properties, you can combine them in 2d array. Maybe, I do not understand the question.
Best Regards,
Pavel
---
### Author:  [@woodscn](https://discourse.paraview.org/u/woodscn)
#### Post date: [January 25, 2022, 3:48pm UTC](https://discourse.paraview.org/t/select-columns-of-structured-data/8822/3 "2022-01-25T15:48:34Z")
What I want to do is manipulate a single CellData field as if it were a multidimensional, dense, array. It sounds like I can’t do that, though. It wouldn’t be possible with any data type other than a uniform rectilinear grid, anyway.
I expect I can use indexing to work around this, though? Something like:
```auto
nx, ny, nz = field.GetDimensions()
for indx in range(nx):
for indy in range(ny):
new_field[indx, indy] = sum(field[indx*nx*ny + indy*ny:indx*nx*ny + indy*ny + nz])
```
I’m not thrilled about the idea of working out the details of that indexing strategy, though. Maybe it’s as simple as:
`3d_field = ndarray((nx,ny,nz), buffer = field, order='C')`
These approaches are making a lot of assumptions about how vtkImageData is laid out in memory, though.
---
### Author:  [@pavel](https://discourse.paraview.org/u/pavel)
#### Post date: [January 26, 2022, 8:56am UTC](https://discourse.paraview.org/t/select-columns-of-structured-data/8822/4 "2022-01-26T08:56:03Z")
Thanks for the clarification. If I undestand you correctly, you can use `vtk_to_numpy` python module from `vtk.util.numpy_support` to avoid this problem of thinking about “how vtkImageData is laid out in memory”.
---
### Author:  [@pavel](https://discourse.paraview.org/u/pavel)
#### Post date: [January 26, 2022, 9:07am UTC](https://discourse.paraview.org/t/select-columns-of-structured-data/8822/5 "2022-01-26T09:07:15Z")
E.g. if `img` is a an object of `vtkDataImage`, you could use something like this:
```
from vtk.util.numpy_support import vtk_to_numpy
nx, ny, _ = img.GetDimensions()
field_vtk = img.GetCellData().GetScalars()
field_numpy = vtk_to_numpy(field_vtk)
3d_field_numpy = field_numpy.reshape(nx, ny, -1)
```
I’m not sure there are no errors in a code above (I don’t have an example dataset), but hope it helps.
---
### Author:  [@woodscn](https://discourse.paraview.org/u/woodscn)
#### Post date: [February 1, 2022, 12:16am UTC](https://discourse.paraview.org/t/select-columns-of-structured-data/8822/6 "2022-02-01T00:16:29Z")
So this looks like it’s almost working (as a programmable filter), but the end result is about half what it should be. I’m trying to integrate a 3D density field with respect to ‘Z’.
I can’t for the life of me figure out why I’d need a factor of 2, so I’m wondering if I’ve messed up the data structure interface somehow.
```auto
from vtk.util.numpy_support import vtk_to_numpy, numpy_to_vtk
nx, ny, nz = inputs[0].GetDimensions()
dx, dy, dz = inputs[0].GetSpacing()
integrand = vtk_to_numpy(inputs[0].PointData['density']).reshape(nz, ny, nx)
out_numpy = numpy.add.reduce(integrand, axis=0, keepdims=True)*dz
out_vtk = numpy_to_vtk(out_numpy.flatten(), deep=1)
out_vtk.SetName('density_a')
out_vtk.SetNumberOfComponents(1)
out_vtk.SetNumberOfTuples(nx*ny*1)
out = self.GetOutput()
out.SetDimensions(nx, ny, 1)
out.SetOrigin(inputs[0].GetOrigin())
out.SetSpacing(inputs[0].GetSpacing())
out.GetPointData().AddArray(out_vtk)
```
---
### Author:  [@pavel](https://discourse.paraview.org/u/pavel)
#### Post date: [February 1, 2022, 5:28am UTC](https://discourse.paraview.org/t/select-columns-of-structured-data/8822/7 "2022-02-01T05:28:24Z")
Could you please clarify:
> [@woodscn](#):
>
> why I’d need a factor of 2
Could you share some data?
Edited: code looks pretty much clear. Could you please compare the sum of all initial point values (vtk\_to\_numpy(inputs[0].PointData[‘density’]) multiplied by dz with the sum of all elements of out\_vtk array in the end? Are they the same or is factor of 2 needed?
Edited 2: code from your previous message works fine for my data. Integration is ok. No need for factor of 2.