Repository navigation
[Bug]: Logarithmic scale of the y-axis does not work when the range is small #29544
Closed as not planned
Description
Bug summary
I have a decreasing function $y\colon [1,3600] \to \mathbb{R}^+$.
I try to set the logarithmic scale for the y-axis, but if the range of the function is small (i.e., the initial and final value are closeby), then the logarithmic scale does not work and it only prints the linear scale.
Code for reproduction
import random
import matplotlib.pyplot as plt
T = 3600
n = 1000
y_start = 10**4
y_end = 0.9*y_start
random.seed(42)
# Generate n increasing random values between 1 and T
x = [random.randrange(1, T) for _ in range(n)]
x.sort()
# Generate n decreasing random values between y_start and y_end
y = [random.uniform(y_end, y_start) for _ in x]
y.sort(reverse=True)
fig, ax = plt.subplots()
ax.set_yscale('log')
ax.plot(x, y)
plt.savefig('step_function.pdf')Actual outcome
Expected outcome
I expected the logarithmis scale. In this example it happens if you set y_end = 0.2*y_start
Additional information
No response
Operating system
Linux debian 6.10.11+bpo-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.10.11-1~bpo12+1 (2024-10-03) x86_64 GNU/Linux
Matplotlib Version
3.10.0
Matplotlib Backend
qtagg
Python version
Python 3.13.0
Jupyter version
No response
Installation
conda
Activity
ksunden commented on Jan 30, 2025
When the deviation is sufficiently small, a logarithmic scale does look linear, that just is how the scale works. If you do not have data that scales multiple orders of magnitude, then a log scale is not likely what you want.
As far as I can tell, this is working as expected.
When you are selecting values between 0.9 and 1.0 times your power of 10, as in this example, you are using only ~5% of one decade.
When you change that to 0.2 to 1.0 times your power of 10, that increases to ~70%, which is enough to see the deviations. (Though still not actually spanning multiple orders of magnitude, where logarithmic really shines).