python Matplotlib tight_layout() never work well

Question

Summary: I want to use Matplotlib.tight_layout() to automatically optimize the layout of figures that contains any number of subfigures with 2D numpy.array.

Example: Here is an example (the data can be accessed from http://www.filedropper.com/fmap). I want to plot the feature map of an CNN layer (later I want to make this automatic, so the code can plot feature maps of any layer from any model). Here I only show the code for one layer for demonstration:

fmap.shape # (1, 64, 64, 128)
square1 = int(round(math.sqrt(fmap.shape[-1])/8)*8) # 8
square2 = int(fmap.shape[-1]/square1) # 16

ix = 1
for _ in range(square1):
    for _ in range(square2):
        # specify subplot and turn of axis
        ax = plt.subplot(square1, square2, ix)
        ax.set_xticks([])
        ax.set_yticks([])
        # plot filter channel in grayscale
        plt.imshow(fmap[0, :, :, ix-1], cmap='gray')
        ix += 1

# show the figure
plt.tight_layout()
plt.show()

The plot is shown as: The space between subfigures and the margin are quite random, as can be seen from feature maps of different layers (the number of subplot differs):

I suggest reviewing matplotlib: Tight Layout guide & matplotlib.pyplot.tight_layout. Try using the rect parameter. It's expected that the padding is dependant upon the number of subplots and the figure size and will be different depending on the configuration. — Trenton McKinney
– Trenton McKinney, Commented Jul 28, 2021 at 20:25
Thanks for the suggestions. I have been playing with those parameters quite a lot. But there is no universal way to make them work. Please help if you have more experience on how to choose these parameters. The things I have tried include, h_adjust, w_adjust, rect, fig_size, etc. — Joey Fueng
– Joey Fueng, Commented Jul 28, 2021 at 21:39
It is also important to note that you are using imshow which defaults to aspect='equal', always showing square images when x and y dimensions are equal. If you really want tight layout, you either need to set a suitable figure width and height, or use imshow(..., aspect='auo') (which stretches the images to fit into the desired layout). — JohanC
– JohanC, Commented Jul 29, 2021 at 10:22

Zephyr · Accepted Answer · 2021-09-06 19:17:48Z

5

You can set figure layout parameters with subplots_adjust.

Basic Code

import matplotlib.pyplot as plt

N = 5

fig, ax = plt.subplots(N, N)

plt.show()

Complete Code

import matplotlib.pyplot as plt

N = 5

fig, ax = plt.subplots(N, N)

plt.subplots_adjust(left = 0.1, top = 0.9, right = 0.9, bottom = 0.1, hspace = 0.5, wspace = 0.5)

plt.show()

To avoid making too many attempts to set the values of left, top, etc..., I suggest you to plot a figure, press the Configure subplots button:

and manually change the values of those six parameters, until you find a configuration that satisfies you. At this point, you can set those six values directly via code using subplots_adjust.

edited Sep 6, 2021 at 19:17

answered Jul 29, 2021 at 9:08

Zephyr

12.6k91 gold badges53 silver badges92 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Joey Fueng Over a year ago

Hi Andrea, thanks for providing a solution. That can only partially address my question, as I have quite some figures to plot and really need it to be fully-automatic. Your answer is a nice workaround, but still needs some trial-and-error.

Jacob Ivanov Apr 29 at 19:46

First time I've ever used this button and I love it. Thanks!

Collectives™ on Stack Overflow

python Matplotlib tight_layout() never work well

1 Answer 1

Basic Code

Complete Code

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Basic Code

Complete Code

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related