"""
=========
Log scale
=========

Examples of plots with logarithmic axes.

You can set the x/y axes to be logarithmic by passing "log" to `~.Axes.set_xscale` /
`~.Axes.set_yscale`.

Convenience functions ``semilogx``, ``semilogy``, and ``loglog``
----------------------------------------------------------------
Since plotting data on semi-logarithmic or double-logarithmic scales is very common,
the functions `~.Axes.semilogx`, `~.Axes.semilogy`, and `~.Axes.loglog` are shortcuts
for setting the scale and plotting data; e.g. ``ax.semilogx(x, y)`` is equivalent to
``ax.set_xscale('log'); ax.plot(x, y)``.
"""

import matplotlib.pyplot as plt
import numpy as np

fig, (ax1, ax2, ax3) = plt.subplots(1, 3, layout='constrained', figsize=(7, 7/3))
# log x axis
t = np.arange(0.01, 10.0, 0.01)
ax1.semilogx(t, np.sin(2 * np.pi * t))
ax1.set(title='semilogx')
ax1.grid()
ax1.grid(which="minor", color="0.9")

# log y axis
x = np.arange(4)
ax2.semilogy(4*x, 10**x, 'o--')
ax2.set(title='semilogy')
ax2.grid()
ax2.grid(which="minor", color="0.9")

# log x and y axis
x = np.array([1, 10, 100, 1000])
ax3.loglog(x, 5 * x, 'o--')
ax3.set(title='loglog')
ax3.grid()
ax3.grid(which="minor", color="0.9")

# %%
# Logarithms with other bases
# ---------------------------
# By default, the log scale is to the base 10. One can change this via the *base*
# parameter.
fig, ax = plt.subplots()
ax.bar(["L1 cache", "L2 cache", "L3 cache", "RAM", "SSD"],
       [32, 1_000, 32_000, 16_000_000, 512_000_000])
ax.set_yscale('log', base=2)
ax.set_yticks([1, 2**10, 2**20, 2**30], labels=['kB', 'MB', 'GB', 'TB'])
ax.set_title("Typical memory sizes")
ax.yaxis.grid()

# %%
# Dealing with negative values
# ----------------------------
# Non-positive values cannot be displayed on a log scale. The scale has two options
# to handle these. Either mask the values so that they are ignored, or clip them
# to a small positive value. Which one is more suited depends on the type of the
# data and the visualization.
#
# The following example contains errorbars going negative. If we mask these values,
# the bar vanishes, which is not desirable. In contrast, clipping makes the value
# small positive (but well below the used scale) so that the error bar is drawn
# to the edge of the Axes.
x = np.linspace(0.0, 2.0, 10)
y = 10**x
yerr = 1.75 + 0.75*y

fig, (ax1, ax2) = plt.subplots(1, 2, layout="constrained", figsize=(6, 3))
fig.suptitle("errorbars going negative")
ax1.set_yscale("log", nonpositive='mask')
ax1.set_title('nonpositive="mask"')
ax1.errorbar(x, y, yerr=yerr, fmt='o', capsize=5)

ax2.set_yscale("log", nonpositive='clip')
ax2.set_title('nonpositive="clip"')
ax2.errorbar(x, y, yerr=yerr, fmt='o', capsize=5)

plt.show()