[DOC] New figure for the gallery (showcase section)

[rougier]

This is a figure for the showcase section. It shows the different elements composing a figure.

[NelleV]

The two lines can be replaced by the following (which I find slightly more elegant):

fig, ax = plt.subplots(figsize=(8, 8), facecolor="w", subplot_kw={"aspect": 1})

[rougier]

I'm not too fond of this syntax, I find it more confusing.

[jenshnielsen]

I think both ways are fine, But in the current I would replace the "matlab like" shorthand syntax

ax = plt.subplot(111, aspect=1)

with the more explicit version

ax = plt.subplot(1, 1, 1, aspect=1)

http://matplotlib.org/api/pyplot_api.html#matplotlib.pyplot.subplot

[jenshnielsen]

Really cool and useful plot. We should perhaps find a way to integrate it into the prose docs too

[tacaswell]

Should replace the image at http://matplotlib.org/faq/usage_faq.html#parts-of-a-figure

[tacaswell]

I think this should be 'Axes`

[tacaswell]

This should be 'Figure'

[rougier]

I've added a commit to change the figure in the FAQ. I canot yet compile the documentation to check the result though.

[tacaswell]

@rougier What is still blocking you on the docs?

[rougier]

Oh only local problems, mostly because matplotlib is already installed from what I've understood.

[NelleV]

I can help with that if you come to see me.

On 9 September 2016 at 05:57, Nicolas P. Rougier notifications@github.com
wrote:

I've added a commit to change the figure in the FAQ. I canot yet compile
the documentation to check the result though.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#7072 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AALR3kIUyWVGagbhEXSnJegIE62UVWAEks5qoVe0gaJpZM4J46gH
.

[Kojoley]

I think all of this could be simplified to one Circle call with help of PathEffects.withStroke http://matplotlib.org/examples/pylab_examples/patheffect_demo.html

[rougier]

Not totally because of the transparency of the background (while the white outline is opaque)

[Kojoley]

Try this, very close to what you have

def circle(x, y, radius=0.15):
    from matplotlib.patheffects import withStroke
    circle = Circle((x, y), radius, clip_on=False, zorder=10, linewidth=1,
                    edgecolor='black', facecolor=(0, 0, 0, .0125),
                    path_effects=[withStroke(linewidth=5, foreground='w')])
    plt.axes().add_artist(circle)

[efiring]

This might further the misconception that markers must be made using scatter rather then plot. Would it be too cluttered it you were to illustrate the scatter plot using markers with varying size, and possibly color, and then show uniform markers in one of the two lines?

[QuLogic]

Have you tried building this figure with the 2.x branch? There are many small tweaks that may have been necessary before, but likely aren't now, e.g., there's no reason to turn off the frame legend now.

Also, the Figure and Axes annotations are a bit vague.

[Kojoley]

@rougier I have opened rougier#2 with circle creation simplification and OO modification.

[rougier]

@Kojoley Thanks, didn't know color arguments accept RGBA format. Your code much simpler now.

[rougier]

@efiring What would be the use case for scatter plot then ? It's seem reasonable to use scatter to plot a bunch of markers, no?

[NelleV]

@efiring @rougier We've had this problem with plot vs scatter since a very very long time. I think that the fact we are having still this problem shows an interface and naming problem (ie, a bug). How complicated it would it be to fix scatter so that it behaves like plot when no size or color argument are provided?

[NelleV]

@rougier the scatter plots is "supposed" to be used when displaying markers of different size or color (or both). It recomputes a bunch of elements for each marker you draw, while plot doesn't. It renders scatter plot quite inefficient both time wise and memory wise when used on large datasets.
There is absolutely no way a user would know this unless he/she had a look at the implementation.

[efiring]

@NelleV Rather than modify scatter to return completely different artists (Line2D versus PatchCollection) depending on its arguments, I think the problem is addressed better via documentation and examples. Using scatter when plot would suffice is inefficient for very large data sets, as you note, but otherwise not harmful.

[efiring]

@NelleV Note also that the argument lists for plot and scatter are very different, corresponding to the major differences between the artists they return.

[rougier]

Is it ready to merge or does it need more modifications ?

[tacaswell]

backported to v2.x as ca51e8d

[QuLogic]

So we're keeping this with the ticks the wrong way?

[QuLogic]

Also, in v2.x, the markers are invisible.