I would like to include a filled contour plot to a pdf document (for example a TeX document).
Currently I am using pyplots contourf, and saving to pdf with pyplots savefig. The problem with this is that the size of the plots becomes rather big as compared to a high resolution png.
One way to reduce the size is of course to reduce the number of levels in the plot, but too few levels gives a poor plot. I'm searching for a simple way to for example let the colors of the plot be saved as a png, with the axes, ticks etc. to be saved vectorized.
You can do this using the Axes option set_rasterization_zorder.
Anything with a zorder less than what you set that to be will be saved as rasterized graphics, even when saving to a vector format like pdf.
For example:
import matplotlib.pyplot as plt
import numpy as np
data = np.random.rand(500,500)
# fig1 will save the contourf as a vector
fig1,ax1 = plt.subplots(1)
ax1.contourf(data)
fig1.savefig('vector.pdf')
# fig2 will save the contourf as a raster
fig2,ax2 = plt.subplots(1)
ax2.contourf(data,zorder=-20)
ax2.set_rasterization_zorder(-10)
fig2.savefig('raster.pdf')
# Show the difference in file size. "os.stat().st_size" gives the file size in bytes.
print os.stat('vector.pdf').st_size
# 15998481
print os.stat('raster.pdf').st_size
# 1186334
You can see this matplotlib example for more background info.
As pointed out by @tcaswell, to rasterise just one artist without having to affect its zorder, you can use .set_rasterized. However, this doesn't appear to be an option with contourf, so you would need to loop over the PathCollections created by contourf and set_rasterized on each of them. Something like this:
contours = ax.contourf(data)
for pathcoll in contours.collections:
pathcoll.set_rasterized(True)
