Why are uint8 arrays casted to booleans? about pyvista HOT 5 CLOSED

You can’t load Boolean arrays directly into vtk objects, so you have to convert them to and from uint8.

from pyvista.

Ahh, I see. Unfortunately, this comes at a price. I had some data arrays from using the vtkTIFFReader that produced a data array of unit8 and this forced it to a boolean when I wrapped the output object:

>>> import vtk
>>> import vtki
>>> from vtk.util.numpy_support import vtk_to_numpy

>>> reader = vtk.vtkTIFFReader()
>>> reader.SetFileName('sample.tif')
>>> reader.Update()
>>> image = vtki.wrap(reader.GetOutputDataObject(0))

>>> image.point_arrays['Tiff Scalars'].dtype
dtype('bool')
>>> arr = image.GetPointData().GetArray('Tiff Scalars')
>>> vtk_to_numpy(arr).dtype
dtype('uint8')

from pyvista.

Ah, I can see why this is an issue. I’ll have to make some changes internally to account for this issue and remove this from the code. You can feel free to remove it if you wish.

from pyvista.

What about vtkBitArray? We could send boolean arrays to this VTK type. I'm thinking to do this we'd need to implement our own vtk_to_numpy and numpy_to_vtk methods in the utilities to check the data types since the numpy_support module in VTK cannot handle boolean/bit data arrays. Here's an example of going back and forth between bit arrays and boolean arrays (I implemented something similar in PVGeo for string arrays which might be good to transfer here as well):

import vtk
from vtk.util.numpy_support import vtk_to_numpy, numpy_to_vtk
import numpy as np

# Create some arbitrary vtkBitArray
arr = vtk.vtkBitArray()
n = 10
arr.SetNumberOfTuples(n)
for i in range(n):
    arr.SetValue(i, i%2)

# Convert the vtkBitArray to NumPy boolean array
iarr = vtk.vtkCharArray()
iarr.DeepCopy(arr)
narr = np.frombuffer(iarr, dtype=bool)

# Convert NumPy boolean array back to a vtkBitArray
carr = numpy_to_vtk(narr.astype(np.uint8))
foo = vtk.vtkBitArray()
foo.DeepCopy(carr)

# Make sure the initial and final arrays are equal:
for i in range(n):
    assert foo.GetValue(i) == arr.GetValue(i)

from pyvista.

This is probably the best approach since numpy doesn't support casting bitarray to uint8 casting, and frombuffer doesn't support it either.

I'm still leaving the old implementation to allow for in-place updating of native np.bool arrays, but I'm adding support for vtkBitArrays and uint8 will remain internally as uint8.

Fixed in dfacba2.

from pyvista.