Option to Compress Graphs for pgf-backend #5983
Description
Working with the awesome .pgf backend, I hit a limit as to how pretty I can make the graphs look:
TeX capacity exceeded, sorry [main memory size={large number here}]
The reason is obvious, when i produce a scatter plot with tens of thousands of objects, it will inevitably produce very large *.pgf documents. Now, I may have overlooked some features and am not an expert on this, but in some cases it may be beneficial if there was an option to save the graph itself as a png and let latex only do the annotations (afterall, its the beautiful typesetting we are after).
As an example my current workaround ("hack") is posted below. Obviously the quality of the graphs is reduced, and the pdf in my example has a bigger filesize, but i have produced *.pgfs of several mb which reduced to some kb in pdf (if they can be compiled at all). In my example the *.pgf size reduces by a third.
I would try to implement this myself in matplotlib, but am not sure where to start.
Is this a good feature, or are there better ways to achieve what I want?
import numpy as np
import matplotlib as mpl
import matplotlib.image as mpimg
mpl.rc('pdf', fonttype=42)
pgf_with_latex = { # setup matplotlib to use latex for output
"pgf.texsystem": "pdflatex", # change this if using xetex or lautex
"text.usetex": True, # use LaTeX to write all text
"font.family": "serif",
}
mpl.rcParams.update(pgf_with_latex)
import matplotlib.pyplot as plt
def newfig(width):
plt.clf()
fig = plt.figure()#figsize=figsize(width))
ax = fig.add_subplot(111)
#ax2 = fig.add_subplot(122)
return fig, ax
def savefig(filename):
plt.savefig('{}.pgf'.format(filename))
plt.savefig('{}.pdf'.format(filename))
np.random.seed(42)
randomData = np.random.rand(50,4)
fig, ax = newfig(0.9)
tmp = ax.scatter(x=randomData[:,0],
y=randomData[:,1]*10,
s=randomData[:,2]*1000,
c=randomData[:,3])
#remember the figure size
frame_ax = ax.get_window_extent().get_points()
frame_fig = fig.get_window_extent().get_points()
#keep the axis labels
ylim = ax.get_ylim()
xlim = ax.get_xlim()
extent = [xlim[0],xlim[1],ylim[0],ylim[1]]
#ax.annotate('help',xy=[0.5,0.5],xytext=[0.4,0.4]) #annotating has to be done after
savefig('large')
ax.axis('off')
plt.savefig('empty.png')
ax.axis('on')
tmp.remove() #remove the graphs
img = mpimg.imread('empty.png')
#print(img.shape)
#get only the figure part
x1 = (int(round(frame_ax[0][0]/frame_fig[1][0]*img.shape[1])))
x2 = (int(round(frame_ax[1][0]/frame_fig[1][0]*img.shape[1])))
y1 = (int(round(frame_ax[0][1]/frame_fig[1][1]*img.shape[0])))
y2 = (int(round(frame_ax[1][1]/frame_fig[1][1]*img.shape[0])))
img2 = img[y1:y2,x1:x2] #crop image at figure
ax.imshow(img2,extent=extent, aspect='auto')#,extent=extent,aspect='auto')
savefig('small')
plt.show()Notes:
In this code I plot, save an image of just the plot without the axis as 'empty.png';
The problem is that this has the size of the figure and not the graph itself, so i have to crop the image.
This cropped image is then included using .imshow.
I looked in the source and knew that the pgf backend has a draw_image function, but got lost trying to figure out how it determines whether the plotted stuff is an image or not.
I may have confused my x's and y's.