Python Headless MatplotLib / Pyplot

Question

I'm trying to make my data analysis and reports less eye stabbing and more graphical with automatically generated graph-files, and to do this I've been playing with matplotlib/pyplot/pylab. Works brilliantly, but when I try to run it on a headless server...

tkinter.TclError: no display name and no $DISPLAY environment variable

For this application I only use PyLab, but after a little google, I added the below to before the pylab import:

import matplotlib
matplotlib.use("Agg")

Which should have changed the backend, but to no effect.

Any ideas what I'm doing wrong?

The remote machine does have X-forwarding capabilities, but since this application shouldn't NEED to display anything, I believe the usual ssh -X hack is overkill.

Examples:Fiendish Deception

Example working code on same machine

import matplotlib
matplotlib.use("Agg")
import numpy as np
import pylab as pl


xvals=np.arange(100)
yvals=np.cumsum(np.random.random(100))
yvals[-10:]=0
yvals=np.log(yvals)
pl.close()

pl.plot(xvals,yvals)
pl.xlabel("X")
pl.ylabel("Y")
pl.title("Title")

pl.savefig("testgraph.png")

Non-working real code

import matplotlib
matplotlib.use("Agg")
import numpy as np
import pylab as pl
import utility as util
import os

... non graph stuff...
def graph_p(self):
    pl.close()
    channels=range(self.p.shape[0])
    for line in range(self.p.shape[1]):
        yvals=np.ma.masked_invalid(map(util.watts_to_dbmhz,self.p[:,line]))
        pl.plot(channels,yvals) #Error says it occurs here
    pl.xlabel("Subchannel Index")
    pl.ylabel("Power (dbmhz)")
    pl.title("Plot of per-tone power assignments for %d lines"%self.p.shape[1])
    pl.savefig(self.dest+self.scenario+'-power.png')

Nothing else touches pyplot.

It is very strange, indeed. Can you post a minimum (non-)working example? The `Agg` backend should not need X. — Andrea Spadaccini, Mar 31 '11 at 17:11

score 28 · Answer 1 · answered Mar 31 '11 at 17:31

28

Everything you describe sounds correct. What happens when you run this:

import matplotlib
matplotlib.use('Agg')
import pylab
pylab.plot([1,2], [3,4], linestyle='-')
pylab.savefig('foo.png')

In my environment it produces this (I scaled it down):

enter image description here

answered Mar 31 '11 at 17:31

samplebias

37,113
6
107
103

score 14 · Answer 2 · edited Jan 30 '18 at 20:07

14

Try importing matplotlib and setting the Agg backend before importing numpy:

import matplotlib
matplotlib.use("Agg")

edited Jan 30 '18 at 20:07

Ross Rogers

23,523
27
108
164

answered Mar 31 '11 at 17:34

Andrea Spadaccini

12,378
5
40
54

No change in behaviour. Even when mpl is the first thing imported and 'used' – Bolster Mar 31 '11 at 17:44

score 12 · Accepted Answer · edited Apr 25 '18 at 20:28

12

Turns out a utility file (not mine!) was pulling in pylab for something else. Shifted the matplotlib backend selection into the initial page.

edited Apr 25 '18 at 20:28

ChrisGPT was on strike

127,765
105
273
257

answered Mar 31 '11 at 18:08

Bolster

7,460
13
61
96

6

This seems to do the trick. Import matplotlib **before** anything else that might import pylab. – Cerin May 16 '12 at 15:09

Python Headless MatplotLib / Pyplot

3 Answers3

Linked