Changing the legend fontsize "hides" dotted lines in the legend

[dawehner]

When you change the fontsize of legends using matplotlib.rcParams['legend.fontsize'] you get some kind of crippled result bag. The line in the legend is not shown 100%

http://nbviewer.ipython.org/76ad2a8053ef0b01a7bd

It seems to be that the font properties are somehow applied to the line itself.

[dawehner]

You can fix it by using numpoints=1 as part of the legend call but this should not happen in the first place if you ask me.

[tacaswell]

I don't see what is wrong.

Please post code + images directly on github, the link to your notebook will rot making this report useless.

[dawehner]

Well, the URL encodes the unique ID of the gist, so this notebook with rot much like github could do.

The plotting code:

x = np.linspace(0.0, 2*np.pi, 200)
y = np.sin(x)
y2 = np.cos(x)

plt.plot(x, y, '--', label='Sin')
plt.plot(x, y2, '-.', label='Sin')

plt.legend()

One time with

matplotlib.rcParams['legend.fontsize'] = 'large' # the default value

and once with

matplotlib.rcParams['legend.fontsize'] = 9

Here are the pictures from the notebook.

[dawehner]

The difference is in the legend itself. The second example does somehow not show the full line, so "--" is kind of crippled towards "-."

[tacaswell]

The way that the legend is drawn is it adds another Line2D artist (called the handle) with the same linestyle as the line you want to add to the legend. I suspect what is going on is that the length of the artist in the legend scales with the font size so as you make the mark get shorter. At some point it gets too short to capture a full dash cycle.

The length of the handle is controlled by the rcparam legend.handlelength (or the handlelength kwargs to legend()) . The length is specified as a multiple of the font size (hency why it scales) and the default value is tuned to look good with the default font. If you want to use a smaller font just make handlelength longer:

x = np.linspace(0.0, 2*np.pi, 200)
y = np.sin(x)
y2 = np.cos(x)
fig, ax = plt.subplots()
ax.plot(x, y, '--', label='Sin')
ax.plot(x, y2, '-.', label='cos')

ax.legend(fontsize=9, handlelength=3)

[dawehner]

As written before, there are workarounds, though the question is whether matplotlib could do anything to prevent this from happening.

[tacaswell]

I disagree that what I suggested is a 'workaround', in all cases the library is doing exactly what you are (implicitly through rcparams) telling it to do. That is, all of the code in behaving as-advertised so there is no bug.

Currently the legend code does not make any attempt introspect the artists being used as handles to guess at what 'good' value are.

[dawehner]

Good point. Thank you for the detailed explanation.

On Mon, Jun 30, 2014 at 3:00 PM, Thomas A Caswell notifications@github.com
wrote:

I disagree that what I suggested is a 'workaround', in all cases the
library is doing exactly what you are (implicitly through rcparams)
telling it to do. That is, all of the code in behaving as-advertised so
there is no bug.

Currently the legend code does not make any attempt introspect the artists
being used as handles to guess at what 'good' value are.

—
Reply to this email directly or view it on GitHub
#3171 (comment)
.

[WeatherGod]

I would like to note that even I have never notice the handlelength feature
before. I am of the belief that most "bugs" are really due to the lack of
good documentation. The rest of them are usually from bad design choices
(he mutters to himself while still wrestling with refactoring color
cycling...)

And no, just simply presenting the attributes in a table isn't
necessarially good documentation...

On Mon, Jun 30, 2014 at 9:18 AM, Daniel Wehner notifications@github.com
wrote:

Good point. Thank you for the detailed explanation.

On Mon, Jun 30, 2014 at 3:00 PM, Thomas A Caswell <
notifications@github.com>
wrote:

I disagree that what I suggested is a 'workaround', in all cases the
library is doing exactly what you are (implicitly through rcparams)
telling it to do. That is, all of the code in behaving as-advertised so
there is no bug.

Currently the legend code does not make any attempt introspect the
artists
being used as handles to guess at what 'good' value are.

—
Reply to this email directly or view it on GitHub
<
https://github.com/matplotlib/matplotlib/issues/3171#issuecomment-47528184>

.

—
Reply to this email directly or view it on GitHub
#3171 (comment)
.

[dawehner]

Well yeah as a user of matplotlib I love how simple things can be done really easy but it is kind of sad how many details you have to know to get done slightly more complex things.

[pmreyes2]

I wanted to point out, that when using LineCollection, there is a mismatch of the size of the lines in the plot and the lines of the handle in the legend. Since the lines of the handle are larger than those from the plot, handlelength needs to be used, but the problem is that the lines in the legend are still larger. See this example:

lines1 = [[(0, .5), (.5, 1)], [(.3, .6), (.2, .2)]]
lines2 = [[[0.7, .2], [.8, .4]], [[.5, .7], [.6, .1]]]
lc1 = matplotlib.collections.LineCollection(lines1, linestyles="--")
lc2 = matplotlib.collections.LineCollection(lines2, linestyles="-.")
fig, ax = plt.subplots()
ax.add_collection(lc1)
ax.add_collection(lc2)
ax.legend([lc1,lc2],["line1","line2"],handlelength=3)

[dopplershift]

@pmreyes2 Are you talking about the thickness of the lines? If so, does the following (running against matplotlib 2.0rc2) look better?