Set Matplotlib colorbar size to match graph

Question

I cannot get the colorbar on imshow graphs like this one to be the same height as the graph, short of using Photoshop after the fact. How do I get the heights to match?

Answer 1

The answers provided so far are very helpful, but they may have shortcomings depending on the situation. For example, they:

1. can take up axis space (1, 2, 3), which becomes noticeable when using multiple subplots,
2. can be tricky to tune to find the specific values that best suit each plot (2, 3) or
3. can lead to results that are hard to predict after calling fig.tight_layout() (4).

An alternative that avoids the above shortcomings is to add an inset/child axis via ax.inset_axes:

import matplotlib.pyplot as plt
import numpy as np


x = np.linspace(0, 1, 70)
y = np.linspace(0, 1, 100)
data = (np.stack(np.meshgrid(x, y)) ** 2).sum(0)


fig, axs = plt.subplots(1, 3, figsize=(6, 3))
for ax in axs:
    hm = ax.imshow(data)

# specify lower-left corner, width and height of the colorbar, 
# in coordinates relative to its parent axis
cax = axs[1].inset_axes((1.05, 0, 0.08, 1.0))
fig.colorbar(hm, cax=cax)
fig.suptitle(r"$\bf{Add~axis~with~relative~(axis)~coordinates}$", y=0.87)
fig.tight_layout()

With this alternative, the coordinates and size of the colorbar are relative to its parent axis, ensuring predictable results regardless of the parent axis's shape or the arrangement of other subplots. Result:

These are the results obtained with the most upvoted answers so far (code to reproduce is at the end).

1. Using an axes divider:

2. Using fraction + pad arguments

3. Adding a new axis in Figure coordinates

4. Using shrink + aspect arguments

Code to reproduce:

import matplotlib.pyplot as plt
import numpy as np
from mpl_toolkits.axes_grid1 import make_axes_locatable

plt.close("all")

x = np.linspace(0, 1, 70)
y = np.linspace(0, 1, 100)
data = (np.stack(np.meshgrid(x, y)) ** 2).sum(0)

# using Axes divider
fig, axs = plt.subplots(1, 3, figsize=(6, 3))
for ax in axs:
    hm = ax.imshow(data)
divider = make_axes_locatable(axs[1])
cax = divider.append_axes("right", size="5%", pad=0.05)
fig.colorbar(hm, cax=cax)
fig.suptitle(r"$\bf{Axes~divider}$" + "\nsteals axis space", y=0.92)
fig.tight_layout()

# using fraction + pad
fig, axs = plt.subplots(1, 3, figsize=(6, 3))
for ax in axs:
    hm = ax.imshow(data)
fig.colorbar(hm, ax=axs[1], fraction=0.046, pad=0.04)
fig.suptitle(
    r"$\bf{fraction+pad}$" + "\nsteals axis space + trial and error positioning", y=0.92
)
fig.tight_layout()

# adding axis in absolute (figure) coordinates
fig, axs = plt.subplots(1, 3, figsize=(6, 3))
for ax in axs:
    hm = ax.imshow(data)
cax = fig.add_axes(
    (
        axs[1].get_position().x1 + 0.01,
        axs[1].get_position().y0,
        0.02,
        axs[1].get_position().y1 - axs[1].get_position().y0,
    ),
)
fig.colorbar(hm, cax=cax)
fig.suptitle(
    r"$\bf{Add~axis~with~Figure~coordinates}$"
    + "\nTrial and error + hard to predict after tight_layout",
    y=1.02,
)
fig.tight_layout()

# using shrink=0.7, aspect=20*0.7
fig, axs = plt.subplots(1, 3, figsize=(6, 3))
for ax in axs:
    hm = ax.imshow(data)
fig.colorbar(hm, ax=axs[1], shrink=0.7, aspect=20 * 0.7)
fig.suptitle(
    r"$\bf{shrink=0.7,~aspect=20*0.7}$" + "\nsteals axis space + trial and error",
    y=0.92,
)
fig.tight_layout()

# adding axis in relative (axis) coordinates
fig, axs = plt.subplots(1, 3, figsize=(6, 3))
for ax in axs:
    hm = ax.imshow(data)
cax = axs[1].inset_axes((1.05, 0, 0.08, 1.0))
fig.colorbar(hm, cax=cax)
fig.suptitle(r"$\bf{Add~axis~with~relative~(axis)~coordinates}$", y=0.87)
fig.tight_layout()

Answer 2

The best fix to this problem I found is explained in detail in this page.

Basically, once you initiate figure:

fig, myplot = plt.subplots((1,1), figsize = (12,10), layout = 'constrained')

supplying the argument:

layout = 'constrained'

worked for me.

Answer 3

For these types of plots I like the ImageGrid API from mpl_toolkits.axes_grid1. It's designed for managing multiple fixed aspect plots, but works just fine for a single image.

from matplotlib import pyplot as plt
from mpl_toolkits.axes_grid1 import ImageGrid


fig = plt.figure()
plot = ImageGrid(fig, 111, (1, 1),
                 cbar_mode='single',
                 cbar_location='right',
                 cbar_size='3%',
                 cbar_pad='5%')
im = plot[0].imshow(np.random.randn(2**4, 2**6))
cbar = fig.colorbar(im, cax=plot.cbar_axes[0])

Answer 4

axes_grid1.axes_divider is the prescribed method for this task (matplotlib even has a demo) but by adding the colorbar, it makes the image smaller. If you want to retain the original image size, then the following offers one way (based on Fei Yao's answer).

data = [(1,2,3,4,5),(4,5,6,7,8),(7,8,9,10,11)]

im = plt.imshow(data, cmap='RdBu')
l, b, w, h = plt.gca().get_position().bounds
cax = plt.gcf().add_axes([l + w + 0.03, b, 0.03, h])
plt.colorbar(im, cax=cax)

A convenient function wrapper.

import matplotlib.pyplot as plt

def add_colorbar(im, width=None, pad=None, **kwargs):

    l, b, w, h = im.axes.get_position().bounds       # get boundaries
    width = width or 0.1 * w                         # get width of the colorbar
    pad = pad or width                               # get pad between im and cbar
    fig = im.axes.figure                             # get figure of image
    cax = fig.add_axes([l + w + pad, b, width, h])   # define cbar Axes
    return fig.colorbar(im, cax=cax, **kwargs)       # draw cbar
    

data = [(1,2,3,4,5),(4,5,6,7,8),(7,8,9,10,11)]

# an example usage
im = plt.imshow(data, cmap='RdBu')
add_colorbar(im)

Answer 5

I encountered this problem recently, I used ax.twinx() to solve it. For example:

from matplotlib import pyplot as plt

# Some other code you've written
...

# Your data generation goes here
xdata = ...
ydata = ...
colordata = function(xdata, ydata)

# Your plotting stuff begins here
fig, ax = plt.subplots(1)
im = ax.scatterplot(xdata, ydata, c=colordata)

# Create a new axis which will be the parent for the colour bar
# Note that this solution is independent of the 'fig' object
ax2 = ax.twinx()
ax2.tick_params(which="both", right=False, labelright=False)

# Add the colour bar itself
plt.colorbar(im, ax=ax2)

# More of your code
...

plt.show()

I found this particularly useful when creating functions that take in matplotlib Axes objects as arguments, draw on them, and return the object because I then don't need to pass in a separate axis I had to generate from the figure object, or pass the figure object itself.

Answer 6

If you don't want to declare another set of axes, the simplest solution I have found is changing the figure size with the figsize call.

In the above example, I would start with

fig = plt.figure(figsize = (12,6))

and then just re-render with different proportions until the colorbar no longer dwarfs the main plot.

Answer 7

You can do this easily with a matplotlib AxisDivider.

The example from the linked page also works without using subplots:

import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1 import make_axes_locatable
import numpy as np
    
plt.figure()
ax = plt.gca()
im = ax.imshow(np.arange(100).reshape((10,10)))
    
# create an axes on the right side of ax. The width of cax will be 5%
# of ax and the padding between cax and ax will be fixed at 0.05 inch.
divider = make_axes_locatable(ax)
cax = divider.append_axes("right", size="5%", pad=0.05)
   
plt.colorbar(im, cax=cax)

Answer 8

An alternative is

shrink=0.7, aspect=20*0.7

shrink scales the height and width, but the aspect argument restores the original width. Default aspect ratio is 20. The 0.7 is empirically determined.

Answer 9

I appreciate all the answers above. However, like some answers and comments pointed out, the axes_grid1 module cannot address GeoAxes, whereas adjusting fraction, pad, shrink, and other similar parameters cannot necessarily give the very precise order, which really bothers me. I believe that giving the colorbar its own axes might be a better solution to address all the issues that have been mentioned.

Code

import matplotlib.pyplot as plt
import numpy as np

fig=plt.figure()
ax = plt.axes()
im = ax.imshow(np.arange(100).reshape((10,10)))

# Create an axes for colorbar. The position of the axes is calculated based on the position of ax.
# You can change 0.01 to adjust the distance between the main image and the colorbar.
# You can change 0.02 to adjust the width of the colorbar.
# This practice is universal for both subplots and GeoAxes.

cax = fig.add_axes([ax.get_position().x1+0.01,ax.get_position().y0,0.02,ax.get_position().height])
plt.colorbar(im, cax=cax) # Similar to fig.colorbar(im, cax = cax)

Result

Later on, I find matplotlib.pyplot.colorbar official documentation also gives ax option, which are existing axes that will provide room for the colorbar. Therefore, it is useful for multiple subplots, see following.

Code

fig, ax = plt.subplots(2,1,figsize=(12,8)) # Caution, figsize will also influence positions.
im1 = ax[0].imshow(np.arange(100).reshape((10,10)), vmin = -100, vmax =100)
im2 = ax[1].imshow(np.arange(-100,0).reshape((10,10)), vmin = -100, vmax =100)
fig.colorbar(im1, ax=ax)

Result

Again, you can also achieve similar effects by specifying cax, a more accurate way from my perspective.

Code

fig, ax = plt.subplots(2,1,figsize=(12,8))
im1 = ax[0].imshow(np.arange(100).reshape((10,10)), vmin = -100, vmax =100)
im2 = ax[1].imshow(np.arange(-100,0).reshape((10,10)), vmin = -100, vmax =100)
cax = fig.add_axes([ax[1].get_position().x1-0.25,ax[1].get_position().y0,0.02,ax[0].get_position().y1-ax[1].get_position().y0])
fig.colorbar(im1, cax=cax)

Result

Answer 10

@bogatron already gave the answer suggested by the matplotlib docs, which produces the right height, but it introduces a different problem. Now the width of the colorbar (as well as the space between colorbar and plot) changes with the width of the plot. In other words, the aspect ratio of the colorbar is not fixed anymore.

To get both the right height and a given aspect ratio, you have to dig a bit deeper into the mysterious axes_grid1 module.

import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1 import make_axes_locatable, axes_size
import numpy as np

aspect = 20
pad_fraction = 0.5

ax = plt.gca()
im = ax.imshow(np.arange(200).reshape((20, 10)))
divider = make_axes_locatable(ax)
width = axes_size.AxesY(ax, aspect=1./aspect)
pad = axes_size.Fraction(pad_fraction, width)
cax = divider.append_axes("right", size=width, pad=pad)
plt.colorbar(im, cax=cax)

Note that this specifies the width of the colorbar w.r.t. the height of the plot (in contrast to the width of the figure, as it was before).

The spacing between colorbar and plot can now be specified as a fraction of the width of the colorbar, which is IMHO a much more meaningful number than a fraction of the figure width.

UPDATE:

I created an IPython notebook on the topic, where I packed the above code into an easily re-usable function:

import matplotlib.pyplot as plt
from mpl_toolkits import axes_grid1

def add_colorbar(im, aspect=20, pad_fraction=0.5, **kwargs):
    """Add a vertical color bar to an image plot."""
    divider = axes_grid1.make_axes_locatable(im.axes)
    width = axes_grid1.axes_size.AxesY(im.axes, aspect=1./aspect)
    pad = axes_grid1.axes_size.Fraction(pad_fraction, width)
    current_ax = plt.gca()
    cax = divider.append_axes("right", size=width, pad=pad)
    plt.sca(current_ax)
    return im.axes.figure.colorbar(im, cax=cax, **kwargs)

It can be used like this:

im = plt.imshow(np.arange(200).reshape((20, 10)))
add_colorbar(im)

Answer 11

All the above solutions are good, but I like @Steve's and @bejota's the best as they do not involve fancy calls and are universal.

By universal I mean that works with any type of axes including GeoAxes. For example, it you have projected axes for mapping:

projection = cartopy.crs.UTM(zone='17N')
ax = plt.axes(projection=projection)
im = ax.imshow(np.arange(200).reshape((20, 10)))

a call to

cax = divider.append_axes("right", size=width, pad=pad)

will fail with: KeyException: map_projection

Therefore, the only universal way of dealing colorbar size with all types of axes is:

ax.colorbar(im, fraction=0.046, pad=0.04)

Work with fraction from 0.035 to 0.046 to get your best size. However, the values for the fraction and paddig will need to be adjusted to get the best fit for your plot and will differ depending if the orientation of the colorbar is in vertical position or horizontal.

Answer 12

This combination (and values near to these) seems to "magically" work for me to keep the colorbar scaled to the plot, no matter what size the display.

plt.colorbar(im,fraction=0.046, pad=0.04)

It also does not require sharing the axis which can get the plot out of square.

Answer 13

When you create the colorbar try using the fraction and/or shrink parameters.

From the documents:

fraction 0.15; fraction of original axes to use for colorbar

shrink 1.0; fraction by which to shrink the colorbar