Seaborn: Too many decimal places in legend labs for relplot with hue or size keywords

[mojones]

I ran into this problem with a particular dataset. Calling replot with either hue or size causes extremely long labels in the legend because of what looks like floating point arithmetic in the labeling logic.

Example:

test = pd.DataFrame({
 'foo' :[1,2,3,4,5,6,7],
 'bar' :[4,6,8,3,5,4,6],
    'baz':[0.4, 0.19, 0.35, 0.22, 0.35, 0.31, 0.09]
}
)
sns.relplot(
data=test.head(7),
x='foo',
y='bar',
size='baz')

Is there some way that the legend labels could be rounded or otherwise formatted to prevent this?

[mojones]

Glancing at other recent issues, I think this is related to https://github.com/mwaskom/seaborn/issues/1653

[mojones]

Workaround for anyone else who finds themselves looking at this:

plot =sns.relplot(
data=test.head(7),
x='foo',
y='bar',
size='baz')

leg = plot._legend
for t in leg.texts:
    # truncate label text to 4 characters
    t.set_text(t.get_text()[:4])

[wwang-chcn]

same problem found on scatterplot
`test = pd.DataFrame({
'foo': [1,1,1,1,1,
2,2,2,2,2,
3,3,3,3,3,],
'bar': ['A', 'B', 'C', 'D', 'E'] * 3,
'baz': [
1.113724, 1.684478, 1.258712, 1.093084, 2.043681,
1.373130, 1.486539, 1.325999, 1.428729, 1.688664,
1.325381, 1.345571, 1.304717, 1.366071, 1.454929]
})

ax = sns.scatterplot(
x='foo',
y='bar',
size='baz',
hue='baz',
palette="YlOrRd",
legend='brief',
data=test)
ax.legend(loc='lower left',
bbox_to_anchor=(1, 0),
ncol=1)`

python: 3.7.3
pandas: 0.25.3
matplotlib: 3.1.0
seaborn: 0.9.0

[asaji]

Problem still exists. Issue with the truncation fix as above is it shifts the legend rightward

[mwaskom]

The specific example at the beginning of this thread now produces the expected result, so if you are having a problem in some other context, you need to provide a reproducible example that demonstrates it.