I am currently trying to reduce the file size of a scatter plot. My code looks like:
plt.scatter(a1,b1) plt.savefig('test.ps') where a1,b1 are arrays of size 400,000 or so, and it gives a file size of 7.8MB.
I have tried adding
plt.rcParams['path.simplify'] = True before this chunk of code, but the file is still 7.8MB. Is this an issue with how it saves as a ".ps" file or another issue?
One approach is to use plot instead of scatter (you can still produce scatter plots using plot by using the 'o' argument), and use the rasterized keyword argument, like so:
import numpy as np import matplotlib.pyplot as plt a1,b1 = np.random.randn(400000,2).T #mock data of similar size to yours plt.plot(a1,b1,'o',rasterized=True) plt.savefig("test.ps") This should significantly reduce the size of the output file. The text and line art will remain vector, only the points are rasterized, so it is a nice compromise.
Depending on what you're looking to achieve, however, it might be better to histogram your data and plot that instead (e.g. pyplot.hist2d or pyplot.hexbin).
You could consider using e.g. hexbin -- I particularly like this when you have a dense collection of points, since it better indicates where your data is concentrated. For example:
import numpy as np import matplotlib.pylab as pl x = np.random.normal(size=40000) y = np.random.normal(size=40000) pl.figure() pl.subplot(121) pl.scatter(x, y) pl.xlim(-4,4) pl.ylim(-4,4) pl.subplot(122) pl.hexbin(x, y, gridsize=40) pl.xlim(-4,4) pl.ylim(-4,4) 
From the left figure, I would have concluded that the distribution of points between x,y = {-3,3} is roughly equal, which clearly is not the case.
(http://matplotlib.org/examples/pylab_examples/hexbin_demo.html)
