Animated plots exported as gifs?

[TomNicholas]

A gif is worth 1000 plots - can we make them via pyvista-xarray?

I saw this pyvista example and immediately thought about plotting xarray data that has a 'time' dimension as an animated 3D plot over time. Can we just wrap similar code to this example inside the pyvista accessor?

cc @jbusecke

[TomNicholas]

Maybe this is already pretty easy - in which case an example showing how to animate something would be awesome!

[banesullivan]

There have been numerous examples of animated plots. Here are a few more, but I recommend also searching the discussion posts for others

• https://docs.pyvista.org/examples/02-plot/gif.html (the one you linked)
• Rendering in background thread pyvista#4006 (reply in thread)
• https://docs.pyvista.org/examples/02-plot/movie.html
• https://docs.pyvista.org/examples/02-plot/moving_cmap.html
• https://docs.pyvista.org/examples/01-filter/collisions.html
• https://docs.pyvista.org/examples/01-filter/image-fft-perlin-noise.html#animate
• https://docs.pyvista.org/examples/02-plot/linked.html
• https://docs.pyvista.org/examples/01-filter/flying_edges.html#animate-barth-sextic
• https://docs.pyvista.org/examples/02-plot/isovalue.html
• https://docs.pyvista.org/examples/02-plot/orbit.html

In most of these examples, you'll see a common pattern:

1. Load data or develop a mechanism to produce the data for various times
2. Plot data in Plotter - configure how to display
3. Open mp4 or gif with Plotter
4. Within a for-loop, make changes to the scene, e.g. the following, and then write the frame
   • modify the scalar field shown
   • translate the mesh through space
   • move the camera around the scene
5. Close out

Easy enough! Though recently, a lot of work has gone into making animations easier and more streamlined with the addition of algorithm support in plotting (an advanced VTK+PyVista feature). Below, I'm going to share two common approaches I take for creating animations of time series data where the mesh structure typically remains constant but the color-mapped field changes through time

Simple: update mesh in place

import numpy as np
import pyvista as pv

x = np.arange(-10, 10, 0.5)
y = np.arange(-10, 10, 0.5)
x, y = np.meshgrid(x, y)
r = np.sqrt(x**2 + y**2)
z = np.sin(r)

grid = pv.StructuredGrid(x, y, z)
grid['Height'] = z.ravel()

nframe = 15
time_space = np.linspace(0, 2 * np.pi, nframe + 1)
subsampled_time_space = time_space[:nframe]

pl = pv.Plotter()
pl.add_mesh(
    grid,
    scalars="Height",
    lighting=False,
    show_edges=True,
    clim=[-1, 1],
)

pl.open_gif('wiggle.gif')
for t in subsampled_time_space:
    z = np.sin(r + t)
    grid.points[:, -1] = z.ravel()  # modifies points
    grid['Height'] = z.ravel()  # modiefies colormapped scalars
    pl.write_frame()

pl.close()

Advanced: develop a custom algorithm

import numpy as np
import pyvista as pv
from vtk.util.vtkAlgorithm import VTKPythonAlgorithmBase
import xarray as xr


class TimeSeries(VTKPythonAlgorithmBase):
    def __init__(self, mesh, data_array, name):
        VTKPythonAlgorithmBase.__init__(
            self, nInputPorts=0, nOutputPorts=1, outputType=mesh.GetClassName(),
        )
        self.mesh = mesh
        self.data_array = data_array
        self.name = name
        self._time_step = 0
        
    @property
    def time_step(self):
        return self._time_step
    
    @time_step.setter
    def time_step(self, t):
        self._time_step = t
        self.Modified()  # critical
    
    def RequestData(self, request, inInfo, outInfo):
        """Perform algorithm execution."""
        try:
            mesh = self.mesh.copy()
            mesh[self.name] = self.data_array[self.time_step]

            out = self.GetOutputData(outInfo, 0)
            out.ShallowCopy(mesh)
        except Exception as e:  # pragma: no cover
            traceback.print_exc()
            raise e
        return 1
    

mesh = pv.Sphere()
nt = 50
time_data = np.random.rand(nt, mesh.n_cells)

alg = TimeSeries(mesh, time_data, 'Field')

pl = pv.Plotter()
pl.add_mesh(
    alg,
    show_edges=True,
)

pl.open_gif('random.gif')
for t in range(nt):
    alg.time_step = t
    pl.write_frame()

pl.close()

PyVista-Xarray has an algorithm already!

Check out https://github.com/pyvista/pyvista-xarray/blob/main/pvxarray/vtk_source.py

Here's an example of how to open a time series dataset with xarray and leverage this algorithm with PyVista:

import pyvista as pv
from pvxarray import PyVistaXarraySource
import xarray as xr

ds = xr.tutorial.load_dataset("air_temperature")
da = ds.air  # Select DataArray

nt = len(da.time)

alg = PyVistaXarraySource(da, x="lon", y="lat", time="time")

pl = pv.Plotter()
pl.add_mesh(alg)
pl.view_xy()

pl.open_gif("air_temperature.gif")
for tstep in range(nt):
    alg.time_index = tstep
    pl.write_frame()

pl.close()

Note that this has 200+ timesteps and creates a 200MB gif, so I used MP4 to make shareable on web

air_temperature.mp4

Other examples

Take inspiration from the custom algorithms developed in these code bases

Kale

@brendanjmeade and I have been working with time series simulations and have made this module to abstract the data loading and "algorithm"

Screen.Recording.2023-01-25.at.2.22.25.PM.mov

Pan3D

Pan3D uses PyVista-Xarray and GeoVista to create interactive, cartographic, animated plots. This supports time series data and also demonstrates how PyVista-Xarray can load data at varying resolutions

Screen.Recording.2022-09-20.at.12.15.06.PM.mov

[banesullivan]

I've been brainstorming developing a "common/base algorithm" that users can extend to further streamline animations with algorithms... would love to gather feedback on how people are using this though

[banesullivan]

tldr; use from pvxarray import PyVistaXarraySource