Backend: datashader

The datashader package is a high-performance graphics pipeline for visualizing very large quantities of data. For further details, please consult the documentation of datashader.

import numpy as np
import pandas as pd
import datashader.transfer_functions as tf
from bewegung import Video

v = Video(width = 480, height = 270, seconds = 1.0)

@v.sequence()
class Foo:

    @v.layer(
        canvas = v.canvas(
            backend = 'datashader',
            x_range = (-9.6, 9.6),
            y_range = (-5.4, 5.4),
        )
    )
    def bar(self, canvas): # a datashader canvas object

        points = 100000
        df = pd.DataFrame(dict(
            x = np.random.normal(size = points),
            y = np.random.normal(size = points),
        ))

        agg = canvas.points(df, 'x', 'y')
        img = tf.shade(agg, cmap = ['lightblue', 'darkblue'], how = 'log')

        return img # a datashader.transfer_functions.Image object

v.reset()
v.render_frame(v.time(0))
Datashader output

Similar to datashader.Canvas, the function call v.canvas(backend = 'datashader') accepts the following additional keyword arguments:

  • plot_width, width of the video by default

  • plot_height, height of the video by default

  • x_range, by default None

  • y_range, by default None

  • x_axis_type, by default 'linear'

  • y_axis_type, by default 'linear'

  • width, mapped to plot_width

  • height, mapped to plot_height

Layer methods are expected to return datashader.transfer_functions.Image objects or, alternatively, Pillow Image objects.

Warning

If a datashader.transfer_functions.Image object is returned, bewegung will mirror the image along the x-axis, i.e. the y-axis will be flipped. This makes the output consistent with Pillow and pycairo, were the y-axis is positive downwards. The flip can be avoided by manually converting the image to a Pillow Image object before returning, i.e. return img.as_pil() in the above example.