The test failure is unrelated. In /Users/efiring/work/programs/py/mpl/tests/result_images/test_axes_grid1/ there has been a change in the size of the test figure that apparently hasn't been propagated.

Can you post an example image?

The first example below is with this PR in effect, and the second is without it.

Another question about legend defaults: presently scatterpoints is 3 and numpoints is 1. Is there any good reason why scatterpoints should not also be 1?

If the size varies then multiple sized markers are shown in the scatter legend, but there is no scale on it so it is not super helpful.

I am 👍 on making scatter points also default to 1

I think this change is an improvement. The large font size in the legend is overkill and the legend border isn't needed. Only question, what does this look like when the legend box starts to sit on top of data?

I definitely prefer having 1 marker in the legend.

I think the best general result is achieved with something like this:

import numpy as np
import matplotlib.pyplot as plt

x = np.linspace(0, 10, 200)

fig, ax = plt.subplots()
ax.plot(x, np.sin(x), label='sine (long label to test big box)')
ax.plot(x, np.cos(x), label='cosine')
ax.plot(x, (x/10)**2, label='quadratic')
ax.plot(x, (x/10)**3, label='cubic')
ax.set_xlabel('This is the x-axis')
ax.set_ylabel('This is the y-axis')
ax.set_title('Example of a proposed default legend')

plt.rcParams['legend.edgecolor'] = 'none'
plt.rcParams['legend.framealpha'] = 0.8
plt.rcParams['legend.frameon'] = True

ax.legend()

fig.savefig('legend_new_long.png')
plt.show()

which yields

The argument against this is that the PS backend doesn't support alpha. In this case, however, I think that letting that limitation of the archaic PS standard hold us back would be a mistake. The fallback for a PS plot will be no worse than what one would get with a likely alternative: an opaque legend background.

Nice! I like using alpha for this and wouldn't worry about PS not
supporting it.

On Tue, May 17, 2016 at 12:15 PM, Eric Firing [email protected]
wrote:

I think the best general result is achieved with something like this:

import numpy as npimport matplotlib.pyplot as plt

x = np.linspace(0, 10, 200)

fig, ax = plt.subplots()
ax.plot(x, np.sin(x), label='sine (long label to test big box)')
ax.plot(x, np.cos(x), label='cosine')
ax.plot(x, (x/10)**2, label='quadratic')
ax.plot(x, (x/10)**3, label='cubic')
ax.set_xlabel('This is the x-axis')
ax.set_ylabel('This is the y-axis')
ax.set_title('Example of a proposed default legend')

plt.rcParams['legend.edgecolor'] = 'none'
plt.rcParams['legend.framealpha'] = 0.8
plt.rcParams['legend.frameon'] = True

ax.legend()

fig.savefig('legend_new_long.png')
plt.show()

which yields
[image: legend_new_long]
https://cloud.githubusercontent.com/assets/85125/15335679/17e904ba-1c0f-11e6-9563-f4aab118e65b.png
The argument against this is that the PS backend doesn't support alpha. In
this case, however, I think that letting that limitation of the archaic PS
standard hold us back would be a mistake. The fallback for a PS plot will
be no worse than what one would get with a likely alternative: an opaque
legend background.

—
You are receiving this because you commented.
Reply to this email directly or view it on GitHub
#6424 (comment)

Brian E. Granger
Associate Professor of Physics and Data Science
Cal Poly State University, San Luis Obispo
@ellisonbg on Twitter and GitHub
[email protected] and [email protected]

Definitely 👍 on the font size change. I'm a little more iffy on dropping the frame, but agree it fits with "less junk by default"; given that, I think the translucent box is a good approach.

+1 for a smaller legend font. I always end up doing it for publication plots.

Removing the border is fine (alternative I've used often : a light gray border instead of black).

Alpha 0.8 looks great.

I actually find it a bit strange looking that the proposed legend fontsize is so much smaller than the tick labels, axis labels, and title. I think I would prefer to change the default legend fontsize to medium rather than small, or alternatively I would prefer to change the default fontsize of all the text objects to one size smaller, with my absolute preference being the latter option.

Here is the version with medium legend.fontsize instead of small:

And this is what happens with medium legend and all text reduced by changing font.size from 12 to 10:

Yes, I like the 10pt font.

Actually, I kinda like that last one

On Fri, May 20, 2016 at 2:19 PM, Eric Firing [email protected]
wrote:

And this is what happens with medium legend and all text reduced by
changing font.size from 12 to 10:
[image: legend_medium-10]
https://cloud.githubusercontent.com/assets/85125/15437568/8976b23e-1e63-11e6-9de6-496b8095730f.png

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub
#6424 (comment)

Worms swarm out of the can...

Selecting default font size, figure size, and dpi is difficult because so much depends on how the resulting plot will be displayed and used, and on what sort of content it will have. There are questions of proportion, and of absolute size. Will the figure be displayed via projector in a presentation? If so, will it occupy the whole slide, or will it be scaled down? Will it go in a tech report, typically formatted in single-column mode, where space is not at a premium? Or is the target a publication, with text in two columns and every figure shrunken to near-invisibility?
I think the defaults should be slanted towards what works in interactive mode on the screen, in informal documents (tech reports and such), and to some degree in presentations and on the web--since a presentation to one's co-workers, as slides or a web page, is often the next step after one's initial data exploration and analysis. In all these cases we want to have something that works with minimal fiddling; defer most fine-tuning to the final publication stage.
My impression is that for such applications, the present 2.0 default font size of 12 pt is a reasonable choice. It matches typical informal document defaults, and it works in a figure on a slide, even with moderate down-scaling. The advantage of 10 pt is that it allows more information to be packed into a figure--especially a multi-axes figure. The tradeoff is that it increases the likelihood of having figures that don't work in projected slides, and that have text that is smaller than the body text in a report.
In trying to set up the defaults for 2.0, we have the additional problem that mpl doesn't have good scaling mechanisms. Changing the font size often requires changing pad values that logically should scale with the font, but don't. And then there is the whole problem of the Axes position, which remains to be fixed to match the present combination of ticks, labels, etc., or any changes in them.

Maybe what we really need here is to have the full set of defaults tuned for the 12 pt case (probably with legend-medium, not small) and then a style that provides the corresponding set of size-related parameters all tuned for the 10 pt case.

👍 to a 12pt and 10pt tuned styles so long as it does not take too much extra work.

The first step is just getting one of the two working. We are closest to the 12-pt version in our present defaults, but it still needs tweaking because of side effects of the outward ticks.

I am going to merge this as-is.

It is clearly an improvement, looks decent (modulo the axes size issues which is being addressed else where).