Here's a bit more detail to expand on Hooked's answer. When I first read that answer, I missed the instruction to call clf() instead of creating a new figure. clf() on its own doesn't help if you then go and create another figure.
Here's a trivial example that causes the warning:
from matplotlib import pyplot as plt, patches
import os
def main():
path = 'figures'
for i in range(21):
_fig, ax = plt.subplots()
x = range(3*i)
y = [n*n for n in x]
ax.add_patch(patches.Rectangle(xy=(i, 1), width=i, height=10))
plt.step(x, y, linewidth=2, where='mid')
figname = 'fig_{}.png'.format(i)
dest = os.path.join(path, figname)
plt.savefig(dest) # write image to file
plt.clf()
print('Done.')
main()
To avoid the warning, I have to pull the call to subplots() outside the loop. In order to keep seeing the rectangles, I need to switch clf() to cla(). That clears the axis without removing the axis itself.
from matplotlib import pyplot as plt, patches
import os
def main():
path = 'figures'
_fig, ax = plt.subplots()
for i in range(21):
x = range(3*i)
y = [n*n for n in x]
ax.add_patch(patches.Rectangle(xy=(i, 1), width=i, height=10))
plt.step(x, y, linewidth=2, where='mid')
figname = 'fig_{}.png'.format(i)
dest = os.path.join(path, figname)
plt.savefig(dest) # write image to file
plt.cla()
print('Done.')
main()
If you're generating plots in batches, you might have to use both cla() and close(). I ran into a problem where a batch could have more than 20 plots without complaining, but it would complain after 20 batches. I fixed that by using cla() after each plot, and close() after each batch.
from matplotlib import pyplot as plt, patches
import os
def main():
for i in range(21):
print('Batch {}'.format(i))
make_plots('figures')
print('Done.')
def make_plots(path):
fig, ax = plt.subplots()
for i in range(21):
x = range(3 * i)
y = [n * n for n in x]
ax.add_patch(patches.Rectangle(xy=(i, 1), width=i, height=10))
plt.step(x, y, linewidth=2, where='mid')
figname = 'fig_{}.png'.format(i)
dest = os.path.join(path, figname)
plt.savefig(dest) # write image to file
plt.cla()
plt.close(fig)
main()
I measured the performance to see if it was worth reusing the figure within a batch, and this little sample program slowed from 41s to 49s (20% slower) when I just called close() after every plot.
