I’m currently evaluating different python plotting libraries. Right now I’m trying matplotlib and I’m quite disappointed with the performance. The following example is modified from SciPy examples and gives me only ~ 8 frames per second!

Any ways of speeding this up or should I pick a different plotting library?

from pylab import *
import time

ion()
fig = figure()
ax1 = fig.add_subplot(611)
ax2 = fig.add_subplot(612)
ax3 = fig.add_subplot(613)
ax4 = fig.add_subplot(614)
ax5 = fig.add_subplot(615)
ax6 = fig.add_subplot(616)

x = arange(0,2*pi,0.01)
y = sin(x)
line1, = ax1.plot(x, y, 'r-')
line2, = ax2.plot(x, y, 'g-')
line3, = ax3.plot(x, y, 'y-')
line4, = ax4.plot(x, y, 'm-')
line5, = ax5.plot(x, y, 'k-')
line6, = ax6.plot(x, y, 'p-')

# turn off interactive plotting - speeds things up by 1 Frame / second
plt.ioff()


tstart = time.time()               # for profiling
for i in arange(1, 200):
    line1.set_ydata(sin(x+i/10.0))  # update the data
    line2.set_ydata(sin(2*x+i/10.0))
    line3.set_ydata(sin(3*x+i/10.0))
    line4.set_ydata(sin(4*x+i/10.0))
    line5.set_ydata(sin(5*x+i/10.0))
    line6.set_ydata(sin(6*x+i/10.0))
    draw()                         # redraw the canvas

print 'FPS:' , 200/(time.time()-tstart)

First off, (though this won’t change the performance at all) consider cleaning up your code, similar to this:

import matplotlib.pyplot as plt
import numpy as np
import time

x = np.arange(0, 2*np.pi, 0.01)
y = np.sin(x)

fig, axes = plt.subplots(nrows=6)
styles = ['r-', 'g-', 'y-', 'm-', 'k-', 'c-']
lines = [ax.plot(x, y, style)[0] for ax, style in zip(axes, styles)]

fig.show()

tstart = time.time()
for i in xrange(1, 20):
    for j, line in enumerate(lines, start=1):
        line.set_ydata(np.sin(j*x + i/10.0))
    fig.canvas.draw()

print 'FPS:' , 20/(time.time()-tstart)

With the above example, I get around 10fps.

Just a quick note, depending on your exact use case, matplotlib may not be a great choice. It’s oriented towards publication-quality figures, not real-time display.

However, there are a lot of things you can do to speed this example up.

There are two main reasons why this is as slow as it is.

1) Calling fig.canvas.draw() redraws everything. It’s your bottleneck. In your case, you don’t need to re-draw things like the axes boundaries, tick labels, etc.

2) In your case, there are a lot of subplots with a lot of tick labels. These take a long time to draw.

Both these can be fixed by using blitting.

To do blitting efficiently, you’ll have to use backend-specific code. In practice, if you’re really worried about smooth animations, you’re usually embedding matplotlib plots in some sort of gui toolkit, anyway, so this isn’t much of an issue.

However, without knowing a bit more about what you’re doing, I can’t help you there.

Nonetheless, there is a gui-neutral way of doing it that is still reasonably fast.

import matplotlib.pyplot as plt
import numpy as np
import time

x = np.arange(0, 2*np.pi, 0.1)
y = np.sin(x)

fig, axes = plt.subplots(nrows=6)

fig.show()

# We need to draw the canvas before we start animating...
fig.canvas.draw()

styles = ['r-', 'g-', 'y-', 'm-', 'k-', 'c-']
def plot(ax, style):
    return ax.plot(x, y, style, animated=True)[0]
lines = [plot(ax, style) for ax, style in zip(axes, styles)]

# Let's capture the background of the figure
backgrounds = [fig.canvas.copy_from_bbox(ax.bbox) for ax in axes]

tstart = time.time()
for i in xrange(1, 2000):
    items = enumerate(zip(lines, axes, backgrounds), start=1)
    for j, (line, ax, background) in items:
        fig.canvas.restore_region(background)
        line.set_ydata(np.sin(j*x + i/10.0))
        ax.draw_artist(line)
        fig.canvas.blit(ax.bbox)

print 'FPS:' , 2000/(time.time()-tstart)

This gives me ~200fps.

To make this a bit more convenient, there’s an animations module in recent versions of matplotlib.

As an example:

import matplotlib.pyplot as plt
import matplotlib.animation as animation
import numpy as np

x = np.arange(0, 2*np.pi, 0.1)
y = np.sin(x)

fig, axes = plt.subplots(nrows=6)

styles = ['r-', 'g-', 'y-', 'm-', 'k-', 'c-']
def plot(ax, style):
    return ax.plot(x, y, style, animated=True)[0]
lines = [plot(ax, style) for ax, style in zip(axes, styles)]

def animate(i):
    for j, line in enumerate(lines, start=1):
        line.set_ydata(np.sin(j*x + i/10.0))
    return lines

# We'd normally specify a reasonable "interval" here...
ani = animation.FuncAnimation(fig, animate, xrange(1, 200), 
                              interval=0, blit=True)
plt.show()

Matplotlib makes great publication-quality graphics, but is not very well optimized for speed. There are a variety of python plotting packages that are designed with speed in mind:

• http://vispy.org
• http://pyqtgraph.org/
• http://docs.enthought.com/chaco/
• http://pyqwt.sourceforge.net/
  [ edit: pyqwt is no longer maintained; the previous maintainer is recommending pyqtgraph ]
• http://code.google.com/p/guiqwt/

To start, Joe Kington’s answer provides very good advice using a gui-neutral approach, and you should definitely take his advice (especially about Blitting) and put it into practice. More info on this approach, read the Matplotlib Cookbook

However, the non-GUI-neutral (GUI-biased?) approach is key to speeding up the plotting. In other words, the backend is extremely important to plot speed.

Put these two lines before you import anything else from matplotlib:

import matplotlib
matplotlib.use('GTKAgg') 

Of course, there are various options to use instead of GTKAgg, but according to the cookbook mentioned before, this was the fastest. See the link about backends for more options.

For the first solution proposed by Joe Kington ( .copy_from_bbox & .draw_artist & canvas.blit), I had to capture the backgrounds after the fig.canvas.draw() line, otherwise the background had no effect and I got the same result as you mentioned. If you put it after the fig.show() it still does not work as proposed by Michael Browne.

So just put the background line after the canvas.draw():

[...]
fig.show()

# We need to draw the canvas before we start animating...
fig.canvas.draw()

# Let's capture the background of the figure
backgrounds = [fig.canvas.copy_from_bbox(ax.bbox) for ax in axes]

This may not apply to many of you, but I’m usually operating my computers under Linux, so by default I save my matplotlib plots as PNG and SVG. This works fine under Linux but is unbearably slow on my Windows 7 installations [MiKTeX under Python(x,y) or Anaconda], so I’ve taken to adding this code, and things work fine over there again:

import platform     # Don't save as SVG if running under Windows.
#
# Plot code goes here.
#
fig.savefig('figure_name.png', dpi = 200)
if platform.system() != 'Windows':
    # In my installations of Windows 7, it takes an inordinate amount of time to save
    # graphs as .svg files, so on that platform I've disabled the call that does so.
    # The first run of a script is still a little slow while everything is loaded in,
    # but execution times of subsequent runs are improved immensely.
    fig.savefig('figure_name.svg')

I’m kind of confused what is going at the backend when I draw plots in matplotlib, tbh, I’m not clear with the hierarchy of plot, axes and figure. I read the documentation and it was helpful but I’m still confused…

The below code draws the same plot in three different ways –

#creating the arrays for testing
x = np.arange(1, 100)
y = np.sqrt(x)
#1st way
plt.plot(x, y)
#2nd way
ax = plt.subplot()
ax.plot(x, y)
#3rd way
figure = plt.figure()
new_plot = figure.add_subplot(111)
new_plot.plot(x, y)

Now my question is –

1. What is the difference between all the three, I mean what is going under the hood when any of the 3 methods are called?

2. Which method should be used when and what are the pros and cons of using any on those?

Method 1

plt.plot(x, y)

This lets you plot just one figure with (x,y) coordinates. If you just want to get one graphic, you can use this way.

Method 2

ax = plt.subplot()
ax.plot(x, y)

This lets you plot one or several figure(s) in the same window. As you write it, you will plot just one figure, but you can make something like this:

fig1, ((ax1, ax2), (ax3, ax4)) = plt.subplots(2, 2)

You will plot 4 figures which are named ax1, ax2, ax3 and ax4 each one but on the same window. This window will be just divided in 4 parts with my example.

Method 3

fig = plt.figure()
new_plot = fig.add_subplot(111)
new_plot.plot(x, y)

I didn’t use it, but you can find documentation.

Example:

import numpy as np
import matplotlib.pyplot as plt

# Method 1 #

x = np.random.rand(10)
y = np.random.rand(10)

figure1 = plt.plot(x,y)

# Method 2 #

x1 = np.random.rand(10)
x2 = np.random.rand(10)
x3 = np.random.rand(10)
x4 = np.random.rand(10)
y1 = np.random.rand(10)
y2 = np.random.rand(10)
y3 = np.random.rand(10)
y4 = np.random.rand(10)

figure2, ((ax1, ax2), (ax3, ax4)) = plt.subplots(2, 2)
ax1.plot(x1,y1)
ax2.plot(x2,y2)
ax3.plot(x3,y3)
ax4.plot(x4,y4)

plt.show()

Other example:

The names of objects

Matplotlib is strongly object oriented and its principal objects are the figure and the axes (I find the name axes a bit misleading, but probably it’s just me).

You can think of the figure as a canvas, of which you typically specify the dimensions and possibly e.g., the background color etc etc. You use the canvas, the figure, essentially in two ways, placing other objects on it (mostly axes, but also text labels etc) and saving its contents with savefig.

You can think of an axes as a sort of Swiss Army knife, a handy object that offers a tool (e.g. .plot, .scatter, .hist etc) for everything, mostly. You can place one, two, … many axes inside a figure using one of many different methods.

The plt interface

The plt procedural interface was originally developed to mimic the MATLAB™ interface but is not really different from the object oriented interface, even if you don’t make a direct reference to the main objects (i.e., a figure and an axes) these objects are automatically instantiated and each plt method is, essentially, translated to a call of one of the methods of the underlying fundamental objects: e.g., a plt.plot() is a hidden_axes.plot and a plt.savefig is a hidden_figure.savefig.

In every moment you can have an handle on these hidden objects using plt.gcf and plt.gca, and this is sometimes necessary when one of the object methods has not been ported to a method in the plt namespace.

I’d like to add that the plt namespace contains also a number of convenience methods to instantiate, in different ways, figure and axes.

Your examples

1st way

plt.plot(x, y)

Here you use only the plt interface, you can only use a single axes in each figure, but this is what you want when you are doing an exploration of your data, a quick recipe that gets the work done…

2nd way

ax = plt.subplot()
ax.plot(x, y)

Here you use a convenience method in the plt namespace to give a name (and a handle) to your axes object, but btw there is also an hidden figure. You can later use the axes object to plot, to make an histogram etc, all things that you can do with the plt interface, but you can also access all its attributes and modify them with greater freedom.

3rd way

figure = plt.figure()
new_plot = figure.add_subplot(111)
new_plot.plot(x, y)

Here you start instantiating a figure using a convenience method in the plt namespace and later you use only the object oriented interface.

It is possible to bypass the plt convenience method (matplotlib.figure.Figure) but you then have to tweak the figure for a better interactive experience (after all, it’s a convenience method).

Personal recommendations

I suggest bare plt.plot, plt.scatter in the context of an interactive session, possibly using IPython with its %matplotlib magic command, and also in the context of an exploratory Jupyter notebook.

On the other hand the object oriented approach, plus a few plt convenience methods, is the way to go

• if you have a permanent issue to solve once for all with a customized arrangement of finely tuned subplots,
• if you want to embed Matplotlib in the UI of a program you write.

There is a large gray area between these extremes and if you ask me what to do I’d just say “It depends”…

I just finished writing code to make a plot using pylab in Python and now I would like to superimpose a grid of 10×10 onto the scatter plot. How do I do that?

My current code is the following:

x = numpy.arange(0, 1, 0.05)
y = numpy.power(x, 2)

fig = plt.figure()
ax = fig.gca()
ax.set_xticks(numpy.arange(0, 1, 0.1))
ax.set_yticks(numpy.arange(0, 1., 0.1))
plt.scatter(x, y)
plt.show()

And its output is:

What I would like is the following output:

EDIT: Added an exemple, based on Andrey Sobolev’s answer

You want to use pyplot.grid:

x = numpy.arange(0, 1, 0.05)
y = numpy.power(x, 2)

fig = plt.figure()
ax = fig.gca()
ax.set_xticks(numpy.arange(0, 1, 0.1))
ax.set_yticks(numpy.arange(0, 1., 0.1))
plt.scatter(x, y)
plt.grid()
plt.show()

ax.xaxis.grid and ax.yaxis.grid can control grid lines properties.

To show a grid line on every tick, add

plt.grid(True)

For example:

import matplotlib.pyplot as plt

points = [
    (0, 10),
    (10, 20),
    (20, 40),
    (60, 100),
]

x = list(map(lambda x: x[0], points))
y = list(map(lambda x: x[1], points))

plt.scatter(x, y)
plt.grid(True)

plt.show()

In addition, you might want to customize the styling (e.g. solid line instead of dashed line), add:

plt.rc('grid', linestyle="-", color='black')

For example:

import matplotlib.pyplot as plt

points = [
    (0, 10),
    (10, 20),
    (20, 40),
    (60, 100),
]

x = list(map(lambda x: x[0], points))
y = list(map(lambda x: x[1], points))

plt.rc('grid', linestyle="-", color='black')
plt.scatter(x, y)
plt.grid(True)

plt.show()

Using rcParams you can show grid very easily as follows

plt.rcParams['axes.facecolor'] = 'white'
plt.rcParams['axes.edgecolor'] = 'white'
plt.rcParams['axes.grid'] = True
plt.rcParams['grid.alpha'] = 1
plt.rcParams['grid.color'] = "#cccccc"

If grid is not showing even after changing these parameters then use

plt.grid(True)

before calling

plt.show()

The pylab examples page is a very useful source. The example relevant for your question:

http://matplotlib.sourceforge.net/mpl_examples/pylab_examples/scatter_demo2.py http://matplotlib.sourceforge.net/users/screenshots.html#scatter-demo

Here is a small example how to add a matplotlib grid in Gtk3 with Python 2 (not working in Python 3):

#!/usr/bin/env python
#-*- coding: utf-8 -*-

import gi
gi.require_version('Gtk', '3.0')
from gi.repository import Gtk
from matplotlib.figure import Figure
from matplotlib.backends.backend_gtk3agg import FigureCanvasGTK3Agg as FigureCanvas

win = Gtk.Window()
win.connect("delete-event", Gtk.main_quit)
win.set_title("Embedding in GTK3")

f = Figure(figsize=(1, 1), dpi=100)
ax = f.add_subplot(111)
ax.grid()

canvas = FigureCanvas(f)
canvas.set_size_request(400, 400)
win.add(canvas)

win.show_all()
Gtk.main()

I am plotting the same type of information, but for different countries, with multiple subplots with matplotlib. That is, I have 9 plots on a 3×3 grid, all with the same for lines (of course, different values per line).

However, I have not figured out how to put a single legend (since all 9 subplots have the same lines) on the figure just once.

How do I do that?

There is also a nice function get_legend_handles_labels() you can call on the last axis (if you iterate over them) that would collect everything you need from label= arguments:

handles, labels = ax.get_legend_handles_labels()
fig.legend(handles, labels, loc='upper center')

figlegend may be what you’re looking for: http://matplotlib.org/api/pyplot_api.html#matplotlib.pyplot.figlegend

Example here: http://matplotlib.org/examples/pylab_examples/figlegend_demo.html

Another example:

plt.figlegend( lines, labels, loc = 'lower center', ncol=5, labelspacing=0. )

or:

fig.legend( lines, labels, loc = (0.5, 0), ncol=5 )

For the automatic positioning of a single legend in a figure with many axes, like those obtained with subplots(), the following solution works really well:

plt.legend( lines, labels, loc = 'lower center', bbox_to_anchor = (0,-0.1,1,1),
            bbox_transform = plt.gcf().transFigure )

With bbox_to_anchor and bbox_transform=plt.gcf().transFigure you are defining a new bounding box of the size of your figureto be a reference for loc. Using (0,-0.1,1,1) moves this bouding box slightly downwards to prevent the legend to be placed over other artists.

OBS: use this solution AFTER you use fig.set_size_inches() and BEFORE you use fig.tight_layout()

You just have to ask for the legend once, outside of your loop.

For example, in this case I have 4 subplots, with the same lines, and a single legend.

from matplotlib.pyplot import *

ficheiros = ['120318.nc', '120319.nc', '120320.nc', '120321.nc']

fig = figure()
fig.suptitle('concentration profile analysis')

for a in range(len(ficheiros)):
    # dados is here defined
    level = dados.variables['level'][:]

    ax = fig.add_subplot(2,2,a+1)
    xticks(range(8), ['0h','3h','6h','9h','12h','15h','18h','21h']) 
    ax.set_xlabel('time (hours)')
    ax.set_ylabel('CONC ($\mu g. m^{-3}$)')

    for index in range(len(level)):
        conc = dados.variables['CONC'][4:12,index] * 1e9
        ax.plot(conc,label=str(level[index])+'m')

    dados.close()

ax.legend(bbox_to_anchor=(1.05, 0), loc='lower left', borderaxespad=0.)
         # it will place the legend on the outer right-hand side of the last axes

show()

I have noticed that no answer display an image with a single legend referencing many curves in different subplots, so I have to show you one… to make you curious…

Now, you want to look at the code, don’t you?

from numpy import linspace
import matplotlib.pyplot as plt

# Calling the axes.prop_cycle returns an itertoools.cycle

color_cycle = plt.rcParams['axes.prop_cycle']()

# I need some curves to plot

x = linspace(0, 1, 51)
f1 = x*(1-x)   ; lab1 = 'x - x x'
f2 = 0.25-f1   ; lab2 = '1/4 - x + x x' 
f3 = x*x*(1-x) ; lab3 = 'x x - x x x'
f4 = 0.25-f3   ; lab4 = '1/4 - x x + x x x'

# let's plot our curves (note the use of color cycle, otherwise the curves colors in
# the two subplots will be repeated and a single legend becomes difficult to read)
fig, (a13, a24) = plt.subplots(2)

a13.plot(x, f1, label=lab1, **next(color_cycle))
a13.plot(x, f3, label=lab3, **next(color_cycle))
a24.plot(x, f2, label=lab2, **next(color_cycle))
a24.plot(x, f4, label=lab4, **next(color_cycle))

# so far so good, now the trick

lines_labels = [ax.get_legend_handles_labels() for ax in fig.axes]
lines, labels = [sum(lol, []) for lol in zip(*lines_labels)]

# finally we invoke the legend (that you probably would like to customize...)

fig.legend(lines, labels)
plt.show()

The two lines

lines_labels = [ax.get_legend_handles_labels() for ax in fig.axes]
lines, labels = [sum(lol, []) for lol in zip(*lines_labels)]

deserve an explanation — to this aim I have encapsulated the tricky part in a function, just 4 lines of code but heavily commented

def fig_legend(fig, **kwdargs):

    # generate a sequence of tuples, each contains
    #  - a list of handles (lohand) and
    #  - a list of labels (lolbl)
    tuples_lohand_lolbl = (ax.get_legend_handles_labels() for ax in fig.axes)
    # e.g. a figure with two axes, ax0 with two curves, ax1 with one curve
    # yields:   ([ax0h0, ax0h1], [ax0l0, ax0l1]) and ([ax1h0], [ax1l0])
    
    # legend needs a list of handles and a list of labels, 
    # so our first step is to transpose our data,
    # generating two tuples of lists of homogeneous stuff(tolohs), i.e
    # we yield ([ax0h0, ax0h1], [ax1h0]) and ([ax0l0, ax0l1], [ax1l0])
    tolohs = zip(*tuples_lohand_lolbl)

    # finally we need to concatenate the individual lists in the two
    # lists of lists: [ax0h0, ax0h1, ax1h0] and [ax0l0, ax0l1, ax1l0]
    # a possible solution is to sum the sublists - we use unpacking
    handles, labels = (sum(list_of_lists, []) for list_of_lists in tolohs)

    # call fig.legend with the keyword arguments, return the legend object

    return fig.legend(handles, labels, **kwdargs)

PS I recognize that sum(list_of_lists, []) is a really inefficient method to flatten a list of lists but ① I love its compactness, ② usually is a few curves in a few subplots and ③ Matplotlib and efficiency? ;-)

Important Update

If you want to stick with the official Matplotlib API my answer above is perfect, really.

On the other hand, if you don’t mind using a private method of the matplotlib.legend module … it’s really much much much easier

from matplotlib.legend import _get_legend_handles_labels
...

fig.legend(*_get_legend_handles_and_labels(fig.axes), ...)

A complete explanation can be found in the source code of Axes.get_legend_handles_labels in .../matplotlib/axes/_axes.py

While rather late to the game, I’ll give another solution here as this is still one of the first links to show up on google. Using matplotlib 2.2.2, this can be achieved using the gridspec feature. In the example below the aim is to have four subplots arranged in a 2×2 fashion with the legend shown at the bottom. A ‘faux’ axis is created at the bottom to place the legend in a fixed spot. The ‘faux’ axis is then turned off so only the legend shows. Result: https://i.stack.imgur.com/5LUWM.png.

import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec

#Gridspec demo
fig = plt.figure()
fig.set_size_inches(8,9)
fig.set_dpi(100)

rows   = 17 #the larger the number here, the smaller the spacing around the legend
start1 = 0
end1   = int((rows-1)/2)
start2 = end1
end2   = int(rows-1)

gspec = gridspec.GridSpec(ncols=4, nrows=rows)

axes = []
axes.append(fig.add_subplot(gspec[start1:end1,0:2]))
axes.append(fig.add_subplot(gspec[start2:end2,0:2]))
axes.append(fig.add_subplot(gspec[start1:end1,2:4]))
axes.append(fig.add_subplot(gspec[start2:end2,2:4]))
axes.append(fig.add_subplot(gspec[end2,0:4]))

line, = axes[0].plot([0,1],[0,1],'b')           #add some data
axes[-1].legend((line,),('Test',),loc='center') #create legend on bottommost axis
axes[-1].set_axis_off()                         #don't show bottommost axis

fig.tight_layout()
plt.show()

if you are using subplots with bar charts, with different colour for each bar. it may be faster to create the artefacts yourself using mpatches

Say you have four bars with different colours as r m c k you can set the legend as follows

import matplotlib.patches as mpatches
import matplotlib.pyplot as plt
labels = ['Red Bar', 'Magenta Bar', 'Cyan Bar', 'Black Bar']


#####################################
# insert code for the subplots here #
#####################################


# now, create an artist for each color
red_patch = mpatches.Patch(facecolor='r', edgecolor='#000000') #this will create a red bar with black borders, you can leave out edgecolor if you do not want the borders
black_patch = mpatches.Patch(facecolor='k', edgecolor='#000000')
magenta_patch = mpatches.Patch(facecolor='m', edgecolor='#000000')
cyan_patch = mpatches.Patch(facecolor='c', edgecolor='#000000')
fig.legend(handles = [red_patch, magenta_patch, cyan_patch, black_patch],labels=labels,
       loc="center right", 
       borderaxespad=0.1)
plt.subplots_adjust(right=0.85) #adjust the subplot to the right for the legend

This answer is a complement to @Evert’s on the legend position.

My first try on @Evert’s solution failed due to overlaps of the legend and the subplot’s title.

In fact, the overlaps are caused by fig.tight_layout(), which changes the subplots’ layout without considering the figure legend. However, fig.tight_layout() is necessary.

In order to avoid the overlaps, we can tell fig.tight_layout() to leave spaces for the figure’s legend by fig.tight_layout(rect=(0,0,1,0.9)).

Description of tight_layout() parameters.

To build on top of @gboffi’s and Ben Usman’s answer:

In a situation where one has different lines in different subplots with the same color and label, one can do something along the lines of

labels_handles = {
  label: handle for ax in fig.axes for handle, label in zip(*ax.get_legend_handles_labels())
}

fig.legend(
  labels_handles.values(),
  labels_handles.keys(),
  loc="upper center",
  bbox_to_anchor=(0.5, 0),
  bbox_transform=plt.gcf().transFigure,
)