5

I have a large data set that is logarithmic in distribution. I want to make a heat map, so I do a 2D histogram and pass that to implot. Because the data is logarithmic, I am passing the log of the data to the histogram. When I make the plot, however, I want the axis to be restored (ie 10^hist bin values) and log axes. If I set the axis to log style, then the image looks all skewed. The data is already 'logged' from when I passed it to the histogram, so I don't want the image affected, just the axis. So, in the below example, I want the image on the left with the axis on the right.

I guess I could do it with a fake overlayed axis, but I don't like to do that sort of thing if there's a better way...

enter image description here

import numpy as np
import matplotlib.pyplot as plt

x=10**np.random.random(10000)*5
y=10**np.random.random(10000)*5

samps, xedges, yedges = np.histogram2d(np.log10(y), np.log10(x),     bins=50)    

ax = plt.subplot(121)

plt.imshow(samps, extent=[0,5,0,5])
plt.xlabel('Log10 X')
plt.ylabel('Log10 Y')

ax = plt.subplot(122)    
plt.imshow(samps, extent=[10**0,10**5,10**0,10**5])
plt.xlabel('X')
plt.ylabel('Y')
plt.xscale('log')
plt.yscale('log')
plt.show()
Alex
  • 427
  • 6
  • 12

3 Answers3

8

You need to use a custom formatter. Here's an example from the matplotlib docs: https://matplotlib.org/examples/pylab_examples/custom_ticker1.html

I tend to use FuncFormatter as the example does. The main trick is that your function need to take to arguments x and pos. I honestly don't know what pos is for. Perhaps no even intentionally, but you can use FuncFormatter as a decorator, which is what I do below:

%matplotlib inline
import numpy as np
import matplotlib.pyplot as plt

@plt.FuncFormatter
def fake_log(x, pos):
    'The two args are the value and tick position'
    return r'$10^{%d}$' % (x)

x=10**np.random.random(10000)*5
y=10**np.random.random(10000)*5

samps, xedges, yedges = np.histogram2d(np.log10(y), np.log10(x), bins=50)    

fig, (ax1) = plt.subplots()
ax1.imshow(samps, extent=[0, 5, 0, 5])
ax1.xaxis.set_major_formatter(fake_log)
ax1.yaxis.set_major_formatter(fake_log)
ax1.set_xlabel('X')
ax1.set_ylabel('Y')

enter image description here

Paul H
  • 65,268
  • 20
  • 159
  • 136
  • This is useful in that I can just apply it over and over to all the plots instead of having to do a manual edit step on each plot, thanks! – Alex Feb 19 '18 at 14:27
1

If you just want to change labels, you can access these directly by plt.gca().set_xticklabels and plt.gca().set_yticklabels. Here is a simple example changing the _text property of these. 

import numpy as np
import matplotlib.pyplot as plt

x = 10 ** np.random.random(10000) * 5
y = 10 ** np.random.random(10000) * 5

samps, xedges, yedges = np.histogram2d(np.log10(y), np.log10(x), bins=50)

plt.subplot(121)

p = plt.imshow(samps, extent=[0, 5, 0, 5])
plt.xlabel('Log10 X')
plt.ylabel('Log10 Y')

plt.subplot(122)

p = plt.imshow(samps, extent=[0, 5, 0, 5])

# The label handling stuff starts here
plt.pause(0.5)  # Needed to make sure the drawing finished being created
plt.xlabel('X')
plt.ylabel('Y')
plt.draw()

ax = plt.gca()
lbx = ax.get_xticklabels()
lby = ax.get_yticklabels()
for i in range(len(lbx)):
    lbx[i]._text = r'$10^' + lbx[i]._text + '$'
for i in range(len(lby)):
    lby[i]._text = r'$10^' + lby[i]._text + '$'
ax.set_xticklabels(lbx)
ax.set_yticklabels(lby)

plt.show()
Aguy
  • 7,851
  • 5
  • 31
  • 58
0

If you need to plot on top of the image in original coordinates, you can overlay two axes like shown below. Of course you can fine-tune ax[1] to match the original amount of ticks etc.

Example based on this image, see result here.

from matplotlib import pyplot as plt
img = plt.imread("noticks.jpg")
fig, ax = plt.subplots(1,2)
ax[1].set_yscale('log')
ax[1].set_xscale('log')
ax[1].set_xlim((5e-4, 0.2))
ax[1].set_ylim((5e-5,0.05))
ax[1].patch.set_facecolor('none')
ax[0].imshow(img)
ax[1].set_aspect(ax[0].get_aspect())
ax[0].axis('off')
ax[0].set_position(ax[1].get_position())
larssp
  • 31
  • 2