Ticks and labels

An axis has two jobs, and they belong to two different objects. A locator decides where the ticks go. A formatter decides what each one says. Once you know that, tick problems stop being a search and become a question with two possible answers.

import matplotlib.ticker as mticker

ax.yaxis.set_major_locator(mticker.MultipleLocator(20))
ax.yaxis.set_major_formatter(mticker.StrMethodFormatter("{x:,.0f}"))

Note that these hang off ax.xaxis and ax.yaxis, not off ax — this is one of the few places the Axis/Axes distinction from lesson 1 actually shows up in code you write.

Setting them by hand

ax.set_xticks(range(len(items)), items)   # positions, then labels
ax.set_xticks([0, 3, 6])                  # positions only
ax.set_xticks([])                          # remove them entirely

This does both jobs at once, and it's exactly right when you know precisely what you want — seven days, five rating levels, four quarters. Its weakness is that it goes stale: the positions are hardcoded, so the day the data gains a row the labels are wrong and nothing says so.

Rotation

Long category labels collide. The standard fix:

ax.set_xticks(range(len(items)), items, rotation=45, ha="right")

ha="right" is not decoration. Rotating text about its centre leaves it drifting sideways from the tick it belongs to; right-aligning pins the end of the word to the mark. Get this wrong and the labels look subtly, unaccountably off.

The better fix is usually to sidestep it: ax.barh() gives every label a full horizontal line and nothing needs rotating at all.

Locators — a rule instead of a list

• MultipleLocator(20) — a tick every 20.
• MaxNLocator(5) — about five, placed at round numbers. Usually what you want.
• LogLocator() — one per decade, for log axes.
• AutoMinorLocator() — minor ticks between the major ones.

These keep working when the data changes, which hardcoded positions don't.

Formatters

mticker.StrMethodFormatter("{x:,.0f}")     # 30000 -> "30,000"
mticker.PercentFormatter(xmax=1.0)         # 0.256 -> "25.6%"
mticker.FuncFormatter(lambda v, pos: f"₹{v/1e6:.1f}M")

FuncFormatter takes a function of (value, position) and returns a string, which means any formatting you can express in Python is available. Currency, units, ordinals, "2h 30m" — all three lines away.

One specific irritation worth naming: the +1e7 that appears in the corner of an axis with large numbers. That's ScalarFormatter factoring out a common offset. Turn it off with ax.ticklabel_format(useOffset=False), or replace the formatter outright.

Dates

import matplotlib.dates as mdates

ax.xaxis.set_major_locator(mdates.AutoDateLocator())
ax.xaxis.set_major_formatter(mdates.DateFormatter("%d %b"))
fig.autofmt_xdate()      # rotate and right-align, in one call

Dates get their own locator/formatter pair because "a sensible tick" means something different for them — months, quarters, weekdays. Learn fig.autofmt_xdate() before you write the rotation out longhand.

Fewer is better

Ticks are furniture, not data. Five well-spaced ones beat twelve crowded ones. Minor ticks are usually noise. And these two are worth knowing:

ax.tick_params(axis="x", labelsize=9, length=0)   # labels, no marks
ax.margins(x=0)                                   # drop the 5% padding