UVW - Universal VTK Writer

UVW is a small utility library to write XML VTK files from data contained in Numpy arrays. It handles fully-fledged ndarrays defined over {1, 2, 3}-d domains, with arbitrary number of components. There are no constraints on the particular order of components, although copy of data can be avoided if the array is Fortran contiguous, as VTK files are written in Fortran order. UVW supports multi-process writing of VTK files, so that it can be used in an MPI environment.

Getting Started

Here is how to install and use uvw.

Prerequisites

• Python 3. It may work with python 2, but it hasn't been tested.
• Numpy. This code has been tested with Numpy version 1.14.3.
• (Optional) mpi4py only if you wish to use the parallel classes of UVW (i.e. the submodule uvw.parallel)

Installing

This library can be installed with pip:

pip install --user uvw

If you want to activate parallel capabilities, run:

pip install --user uvw[mpi]

which will automatically pull mpi4py as a dependency.

Writing Numpy arrays

As a first example, let us write a multi-component numpy array into a rectilinear grid:

import numpy as np
from uvw import RectilinearGrid, DataArray

# Creating coordinates
x = np.linspace(-0.5, 0.5, 10)
y = np.linspace(-0.5, 0.5, 20)
z = np.linspace(-0.9, 0.9, 30)

# Creating the file (with possible data compression)
grid = RectilinearGrid('grid.vtr', (x, y, z), compression=True)

# A centered ball
x, y, z = np.meshgrid(x, y, z, indexing='ij')
r = np.sqrt(x**2 + y**2 + z**2)
ball = r < 0.3

# Some multi-component multi-dimensional data
data = np.zeros([10, 20, 30, 3, 3])
data[ball, ...] = np.array([[0, 1, 0],
                            [1, 0, 0],
                            [0, 1, 1]])

# Some cell data
cell_data = np.zeros([9, 19, 29])
cell_data[0::2, 0::2, 0::2] = 1

# Adding the point data (see help(DataArray) for more info)
grid.addPointData(DataArray(data, range(3), 'ball'))
# Adding the cell data
grid.addCellData(DataArray(cell_data, range(3), 'checkers'))
grid.write()

UVW also supports writing data on 2D and 1D physical domains, for example:

import sys
import numpy as np
from uvw import RectilinearGrid, DataArray

# Creating coordinates
x = np.linspace(-0.5, 0.5, 10)
y = np.linspace(-0.5, 0.5, 20)

# A centered disk
xx, yy = np.meshgrid(x, y, indexing='ij')
r = np.sqrt(xx**2 + yy**2)
R = 0.3
disk = r < R

data = np.zeros([10, 20])
data[disk] = np.sqrt(1-(r[disk]/R)**2)

# File object can be used as a context manager
# and you can write to stdout!
with RectilinearGrid(sys.stdout, (x, y)) as grid:
  grid.addPointData(DataArray(data, range(2), 'data'))

Writing in parallel with mpi4py

The classes contained in the uvw.parallel submodule support multi-process writing using mpi4py. Here is a code example:

import numpy as np

from mpi4py import MPI

from uvw.parallel import PRectilinearGrid
from uvw import DataArray

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

N = 20

# Domain bounds per rank
bounds = [
    {'x': (-2, 0), 'y': (-2, 0)},
    {'x': (-2, 0), 'y': (0,  2)},
    {'x': (0,  2), 'y': (-2, 2)},
]

# Domain sizes per rank
sizes = [
    {'x': N, 'y': N},
    {'x': N, 'y': N},
    {'x': N, 'y': 2*N-1},  # account for overlap
]

# Size offsets per rank
offsets = [
    [0, 0],
    [0, N],
    [N, 0],
]

x = np.linspace(*bounds[rank]['x'], sizes[rank]['x'])
y = np.linspace(*bounds[rank]['y'], sizes[rank]['y'])

xx, yy = np.meshgrid(x, y, indexing='ij', sparse=True)
r = np.sqrt(xx**2 + yy**2)
data = np.exp(-r**2)

# Indicating rank info with a cell array
proc = np.ones((x.size-1, y.size-1)) * rank

with PRectilinearGrid('pgrid.pvtr', (x, y), offsets[rank]) as rect:
    rect.addPointData(DataArray(data, range(2), 'gaussian'))
    rect.addCellData(DataArray(proc, range(2), 'proc'))

As you can see, using PRectilinearGrid feels just like using RectilinearGrid, except that you need to supply the position of the local grid in the global grid numbering (the offsets[rank] in the above example). Note that RecilinearGrid VTK files need an overlap in point data, hence why the global grid size ends up being (2*N-1, 2*N-1). If you forget that overlap, Paraview (or another VTK-based software) may complain that some parts in the global grid (aka "extents" in VTK) are missing data.

Writing unstructured data

UVW supports VTK's UnstructuredGrid, where the geometry is given with a list of nodes and a connectivity. The UnstructuredGrid class expects connectivity to be a dictionnary enumerating the different connectivity types and the cells associated to each type. For example:

import numpy as np

from uvw import UnstructuredGrid
from uvw.unstructured import CellType

nodes = np.array([
    [0, 0, 0],
    [1, 0, 0],
    [1, 1, 0],
    [0, 1, 0],
    [2, 0, 0],
    [0, 2, 0],
    [1, 2, 0],
])

connectivity = {
    CellType.QUAD: np.array([
        [0, 1, 2, 3], [2, 6, 5, 3],
    ]),
    5: np.array([[4, 2, 1]]),
}

f = UnstructuredGrid('ugrid.vtu', nodes, connectivity)
f.write()

As you can see, cell types can be specified with the unstructured.CellType enumeration or with the underlying integer value (see VTKFileFormats for more info). UnstructuredGrid performs a sanity check of the connectivity to see if the number of nodes matches the cell type.

If you work with large amounts of unstructured data, consider checking out meshio which provides many different read/write capabilities for various unstructured formats, some of which are supported by VTK and are better than VTK's simple XML format.

List of features

Here is a list of what is available in UVW:

VTK file formats

• Image data (.vti)
• Rectilinear grid (.vtr)
• Structured grid (.vts)
• Unstructured grid (.vtu)
• Parallel Rectilinear grid (.pvtr)
• Parallel Image data (.pvti)
• ParaView Data (.pvd)

Data representation

• ASCII
• Base64 (raw and compressed: the compression argument of file constructors can be True, False, or an integer in [-1, 9] for compression levels)

Note that raw binary data, while more space efficient and supported by VTK, is not valid XML, and therefore not supported by UVW, which uses minidom for XML writing.

PyEVTK high-level API implementation

To facilitate transition from PyEVTK, UVW implements a part of its API, without imposing restrictions on data (such as the number of components per array) and allowing data compression. Simply replace import pyevtk.hl by import uvw.dropin.hl. To enable compression, provide compression=True to any of the functions in uvw.dropin.hl. Note: the drop-in is not automatically tested, do not hesitate to report problems.

Planned developments

Here is a list of future developments:

• Image data
• Unstructured grid
• Structured grid
• Parallel writing (mpi4py-enabled PRectilinearGrid and PImageData are now available!)
• Benchmarking + performance comparison with pyevtk

Developing

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Git repository

First clone the git repository:

git clone https://github.com/prs513rosewood/uvw.git

Then you can use pip in development mode (possibly in virtualenv):

pip install --user -e .[mpi,tests]

Installing with the tests extra pulls vtk as a dependency. This is because reading files with VTK in tests is the most reliable way to check file integrity.

Running the tests

The tests can be run using pytest:

pytest
# or for tests with mpi
mpiexec -n 2 pytest --with-mpi

License

This project is licensed under the MIT License - see the LICENSE.md file for details.

Acknowledgments

• @PurpleBooth's README-Template