Thank you very much for suggestions.
Below, I join my own solution. It is hardly "a minimum working example" but I have already stripped my script quite a lot!
In a nutshell, I used imshow to plot the "image" (a 2D histogram with log bins) and I remove the axes. Then, I draw a second, empty (and transparent), plot, exactly on top of the first plot just to get log axes as imshow doesn't seem to allow it. Quite complicated if you ask me!
My code is probably far from optimal as I am new to python and matplotlib...
By the way, I don't use hexbin for two reasons:
1) It is too slow to run on very big data files like the kind I have.
2) With the version I use, the hexagons are slightly too large, i.e. they overlap, resulting in "pixels" of irregular shapes and sizes.
Also, I want to be able to write the histogram data into a file in text format.
#!/usr/bin/python
# How to get log axis with a 2D colormap (i.e. an "image") ??
#############################################################
#############################################################
import numpy as np
import matplotlib.cm as cm
import matplotlib.pyplot as plt
import math
# Data file containing 2D data in log-log coordinates.
# The format of the file is 3 columns : x y v
# where v is the value to plotted for coordinate (x,y)
# x and y are already log values
# For instance, this can be a 2D histogram with log bins.
input_file="histo2d.dat"
# Parameters to set space for the plot ("bounding box")
x1_bb, y1_bb, x2_bb, y2_bb = 0.125, 0.12, 0.8, 0.925
# Parameters to set space for colorbar
cb_fraction=0.15
cb_pad=0.05
# Return unique values from a sorted list, will be required later
def uniq(seq, idfun=None):
# order preserving
if idfun is None:
def idfun(x): return x
seen = {}
result = []
for item in seq:
marker = idfun(item)
# in old Python versions:
# if seen.has_key(marker)
# but in new ones:
if marker in seen: continue
seen[marker] = 1
result.append(item)
return result
# Read data from file. The format of the file is 3 columns : x y v
# where v is the value to plotted for coordinate (x,y)
Data = np.genfromtxt(input_file)
x = Data[:,0]
y = Data[:,1]
v = Data[:,2]
# Determine x and y limits and resolution of data
x_uniq = np.array(uniq(np.sort(x)))
y_uniq = np.array(uniq(np.sort(y)))
x_resolution = x_uniq.size
y_resolution = y_uniq.size
x_interval_length = x_uniq[1]-x_uniq[0]
y_interval_length = y_uniq[1]-y_uniq[0]
xmin = x.min()
xmax = x.max()+0.5*x_interval_length
ymin = y.min()
ymax = y.max()+0.5*y_interval_length
# Reshape 1D data to turn it into a 2D "image"
v = v.reshape([x_resolution, y_resolution])
v = v[:,range(y_resolution-1,-1,-1)].transpose()
# Plot 2D "image"
# ---------------
# I use imshow which only work with linear axes.
# We will have to change the axes later...
axis_lim=[xmin, xmax, ymin, ymax]
fig = plt.figure()
ax = fig.add_subplot(111)
extent = [xmin, xmax, ymin, ymax]
img = plt.imshow(v, extent=extent, interpolation='nearest', cmap=cm.Reds, aspect='auto')
ax.axis(axis_lim)
# Make space for the colorbar
x2_bb_eff = (x2_bb-(cb_fraction+cb_pad)*x1_bb)/(1.0-(cb_fraction+cb_pad))
ax.set_position([x1_bb, y1_bb, x2_bb_eff-x1_bb, y2_bb-y1_bb])
position = ax.get_position()
# Remove axis ticks so that we can put log ticks on top
ax.set_xticks([])
ax.set_yticks([])
# Add colorbar
cb = fig.colorbar(img,fraction=cb_fraction,pad=cb_pad)
cb.set_label('Value [unit]')
# Add logarithmic axes
# --------------------
# Empty plot on top of previous one. Only used to add log axes.
ax = fig.add_subplot(111,frameon=False)
ax.set_xscale('log')
ax.set_yscale('log')
plt.plot([])
ax.set_position([x1_bb, y1_bb, x2_bb-x1_bb, y2_bb-y1_bb])
axis_lim_log=map(lambda x: 10.**x, axis_lim)
ax.axis(axis_lim_log)
plt.grid(b=True, which='major', linewidth=1)
plt.ylabel('Some quantity [unit]')
plt.xlabel('Another quantity [unit]')
plt.show()
