First of all - congrats on your great module perfplot
that uses this style. Although I myself am a great fan of removing chartjunk I guess that less in sometimes more even in trying to do less that is indeed more in many cases.
What I mean is that removing the bottom spine (and left spine if no gridlines are used) and using only sparse tick labels doesn't increase readability, on contrary it can make things much worse:
import dufte
import matplotlib
import matplotlib.pyplot as plt
matplotlib.style.use(dufte.style)
plt.xscale('log')
plt.yscale('log')
plt.plot([15,50,95], [11,90,95])
plt.title('dufte style')
plt.tight_layout()
ย ย ย ย ย ![](https://user-images.githubusercontent.com/19879328/107969601-42c3fd80-6fb0-11eb-8c11-a941b41b8525.png)
I admit this is a specifically constructed minimal example but the reason behind it is that I ended up with such a perfplot where we have just one grid line in the middle of the chart (kind of visually dividing the chart into two halves) and a single x tick label floating in the middle of free space.
.
So my proposal is
- always show the bottom spine (easy)
- adjust the tick locators for log scales (I don't have a universal idea but the goal is to always have more than just one single number which is quite meaningless as you don't have a reference, you can't assess the scale of the whole axis).