Python - Plot a specific subset of large datasets using Holoviews Bokeh

Question

What I am trying to do is to create an interactive plot with Pan and Zoom in/out interactions for large time series.

Consider the next case in Jupyter Notebook:

import numpy as np

import holoviews as hv
import holoviews.plotting.bokeh
from holoviews.operation import decimate

hv.extension('bokeh')

n_samples = 1_000 #100_000_000

x = np.linspace(0.0,10.0, n_samples)

y = np.zeros((64, n_samples))
r = np.random.rand(n_samples)

for i in range(64):
    y[i] = np.sin(r + np.random.rand(n_samples)*0.3)+i


curves = hv.Curve( (zip(x,y[0,:])) ).opts(height=400, width=800)
for i in range(1,64):
    curves *= hv.Curve( (zip(x,y[i,:])) ) 

curves = curves.options({'Curve': {'color': 'black'}})

curves = decimate(curves).collate()

curves.redim(x=hv.Dimension('x', range=(0, 2)))

Using n_samples=1_000 goes well but the current number of samples is ~10-100 million of points, so it goes super slow.

I think this happens because it creates all the graphical elements and stores them in memory. Then, when I change the range in x using the Pan tool, it has to search which of all the elements need to be plotted, and that is the slow part.

If that's the case, a solution may be to plot only a subset of 1k-5k points from the arrays considering the ranges of the canvas. I don't need all the points on the canvas, so they can be computed on the fly.

Do you know another way to solve this issue? I am new using Bokeh and Holoviews.

Thanks!

Answer 1

Sure. As suggested by Sander, you can use Datashader to render your data outside of the browser for speed and efficiency. Once you have defined curves, just do:

import holoviews.operation.datashader as hd

hd.rasterize(curves)  # as the last line in your Jupyter notebook cell

You shouldn't need to decimate.