prs513rosewood / uvw Goto Github PK

Hi,

I'm using uvw successfully to write structured scalar data to vtr-file, and I'm able to open and visualize the data with e.g. Paraview. Now I have vector valued data, e.g. the velocity, which I have stored in three diffrerent numpy arrays for the x,y,z components of the velocity, here U, V, W respectively. I can successfully write the three different ndarrays of shape (nx, ny, nz) as scalar data, and I build vector data from that inside Paraview. I have tried different reshape/ravel/dstack variants (e.g. "velocity=numpy.dstack((U.ravel(), V.ravel(), W.ravel()))"), but I didn't succeed and I have no idea of "how to write the velocity vector, which can easily be build from the components, to the VTK file, such that I can access the vector data in Paraview!?

Please, any help is highly appreciated!

Kind Regards
Jörg

Hi, I'm creating an ImageData file, and when writing the file, the NumberOfTuples property produces the size of the flattened data rather than the number of n-component elements in the data array. For instance

grid = uvw.ImageData('test_vtk.vti', [(-10., 10.), (-5, 5), (-5, 5)], [3, 3, 2], compression=True)
grid.addCellData(uvw.DataArray(np.arange(1., 9.).reshape(-1, 2), spatial_axes=[0], name='x0d'))

Produces a vti file with NumberOfTuples=8 rather than 4 which is is the number of pair data values I set.
Is the NumberOfTuples property necessary? I think the VTK file works as well without this property but it might be needed to allocate memory properly. There is very little documentation of this value in the VTK docs.

I found that the array is flattened here:

Hello,

I think I found a small error in the "unstructured.py"-part. The Hexahedron type should only have 8 nodes instead of 9 in the NODES_PER_CELL list.

CellType.HEXAHEDRON: 8, instead of CellType.HEXAHEDRON: 9, in line 65. This corrects a few errors I had using the .vtu files with paraview such as clipping elements.

Best greetings and thanks - very helpful stuff

Description of problem:

Hi,

Thank you for writing this module, it's far superior to alternatives I have tried with respect to the compression of output.

However, please consult the following example, adapted from the example given in the uvw README:

import numpy as np
from uvw import ImageData, DataArray

x = np.linspace(-0.5, 0.5, 128)
y = np.linspace(-0.5, 0.5, 128)
z = np.linspace(-0.5, 0.5, 128)


cell_data = np.zeros([128, 128, 128])
cell_data[54:74, 54:74, 10:118] = 1.0
cell_data[10:118, 54:74, 54:74] = 1.0
cell_data[54:74, 10:118, 54:74] = 1.0

grid = ImageData(
    "test-good.vti",
    ranges=[(-0.5, 0.5), (-0.5, 0.5), (-0.5, 0.5)],
    points=[128, 128, 128],
    compression=False,
)
grid.addPointData(DataArray(cell_data, range(3), "value"))
grid.write()

grid = ImageData(
    "test-mangled.vti",
    ranges=[(-0.5, 0.5), (-0.5, 0.5), (-0.5, 0.5)],
    points=[128, 128, 128],
    compression=False,
)
grid.addCellData(DataArray(cell_data, range(3), "value"))
grid.write()

This example outputs two VTI files representing two identical ndarrays, distinguished by whether the data are added as a point data array or a cell data array. ParaView renderings of these data are shown:

test-good.vti:

test-mangled.vti:

Note that the data in test-mangled.vti is skewed.

Expected behaviour:

Broadly comparable data representations, distinguished by whether the data is of a cell or point type.

Attempted mitigations:

• Enabling or disabling compression didn't do anything.
• Using a RectilinearGrid produced analogous results, however performance was somewhat impaired in ParaView, likely because the RectilinearGrid structure can be anisotropic.
• Played around with the offset argument of ImageData, no improvement.

Commentary

• This behaviour appears to manifest in the README example, however because the array in the example is populated with a stride-2 grid, the effect is masked.
• This may be a bug, or may be me misunderstanding CellData. If the latter, please accept my apologies and a request for the correct means of populating CellData.
• For the sort of data I am working with, CellData is very much preferred to PointData.

Platform:

uvw 0.5.1
Python 3.10
ParaView 5.11.1
Linux 6.1

If I need to write a file multiple times, is there a way to update the cell or point data appended in a Data object? SO I can write the same VTK but with updated values (and possibly change the filename as well)

Hello,

In the PVTKFile class you set the comm attribute to be self.comm = MPI.COMM_WORLD. Can the user use a communicator other than MPI_COMM_WORLD to write files in parallel ?

When trying to create a rectilinear grid from numpy arrays, I got an error if my arrays were formatted with the data type '>f8': TypeError: Array dtype >f8 is not supported by VTK. This seems similar to an issue recently reported for pyvista (#540).

The following code raised the error:

from uvw import RectilinearGrid, DataArray
import numpy as np

data = np.zeros((5,5,5))

#xdim = np.array([0,1,2,3,4]).astype('>f8')
#ydim = np.array([0,1,2,3,4]).astype('>f8')
#zdim = np.array([0,1,2,3,4]).astype('>f8')

grid = RectilinearGrid('data.vtr', (ydim, xdim, zdim), compression = True)
grid.addPointData(DataArray(data, range(3), 'data'))
grid.write()

Changing xdim, ydim, and zdim as follows resolved the error:

xdim = np.array([0,1,2,3,4], dtype=np.float64)
ydim = np.array([0,1,2,3,4], dtype=np.float64)
zdim = np.array([0,1,2,3,4], dtype=np.float64)

While this may not matter much for cases where the arrays are defined within the code, it can cause problems when data is imported from other files that may have strange formatting.

(Sorry if my terminology is bad, or if this isn't a proper issue. I'm very new to github.)

System info:

OS: Windows 10 Home
Architecture: 64bit

Python 3.7.6
uvw 0.3.1
numpy 1.18.1