Marker style and size issues of legend.

[gepcel]

Let's see an example first:

import ultraplot as pplt
fig,ax=pplt.subplot()
ax.plot(.5, .7, m='.', ms=30, label='plot of size 30')
ax.scatter(.5, .5, ms=30, label='scatter of size 30')
ax.format(xlim=(0, 1), ylim=(0, 1))
ax,legend(loc='ll', ncols=1)
ax.sizelegend([20], labels=['area=True'], loc='lr')
ax.sizelegend([20], labels=['area=False'], loc='ur', area=False)

The result is as following:

I have three questions/suggestions:

1. The marker style of plot method and it's coresponding legend entry (plot of size 30 of this example) is inconsistent.
2. As I understand it, the plot method use size=30 as the diameter. This should theoretically be the same as sizelegend with area=False, which doesn't seem to be the case in the example. Conversely, the scatter method use size=30 as the area, which should match sizelegend with area=True, and this appears to be correct.
3. I suggest add area arg to catlegend, and maybe other semantic legends if you think it's appropriate.

[cvanelteren]

From my understanding the LineCollection and Scatter produce different markersizes by default as you indicated. The default for lines are points and for scatters are area size.

I think this is indeed rather inconsistent from a user perspective. While matplotlib uses the convention above, we should not make the same mistakes here. What we could do is

• ms / markersize always means marker diameter in points
• s always means scatter marker area in points squared (same as mpl)
• sizelegend(area = False) should match ms / markersize
• sizelegend(area=True) should match scatter's s
• catlegend/entrylegend should follow the same rule

This ensures that we still follow matplotlib's convention (and by proxy expectation from a user coming to UltraPlot).

What do you think?

[gepcel]

Would distinguishing between ms/markersize and s lead to confusion in the area=True/False logic? Should s=ms=markersize simply be treated as different aliases for the exact same argument? And use area argument to control if it is area or diameter?

[cvanelteren]

I think that would lead to a break in the expectation one would have when coming from matplotlib. Starting working on this on #746

[cvanelteren]

I mean we could also leave it as is and just adhere to the expectations implicitly from the line / scatter collection @gepcel, don't really have a strong vision here as I have yet to really use these in my work.

[gepcel]

Here is my opinion regarding the legend behavior:

1. The marker style in ax.legend should be consistent with the one used in the corresponding plot. Currently, there is an unexpected inconsistency between the two.
2. For semantic legends, the markersize (including its aliases s and ms) all represent the diameter when area=True, and the area when area=False.
3. Currently, only sizelegend supports the area parameter. I suggest extending this support to all other semantic legends as well.
4. In sizelegend, when area=False, the marker size does not match the marker size used in the plot() method. These two values better be identical.

[cvanelteren]

Thanks! I agree this was inconsistent. The source of confusion is that Matplotlib uses two conventions as I indicated (ms for marker diameter and s for area). UltraPlot made this worse by ccepting overlapping aliases without making the semantics very clear. I have updated the api on #746 such that:

• scatter now follows the plot semantics and is interpreted as diameter
• sizelegend now matches plot semantics
• entrylegend and catlegend now accept area
• For semantic legends, marker-size aliases now obey area: area=False means diameter, area = True means area
• Single-point plot legend entries now hide the connector line so that the legend better matches the rendered artist

The area=True meaning "values are areas" to match the existing sizelegend behavior. I tihnk this interpretation is less surprising and more consistent with the current API.

Tell me what you think!

[gepcel]

The source of confusion is that Matplotlib uses two conventions as I indicated (ms for marker diameter and s for area).

Oh, I didn't know that! Then maybe you can just use the same logic, it's up to you to decide. I've learned something new!

[gepcel]

I have updated the api on #746 such that:

• sizelegend now matches plot semantics

Now, the size of sizelegend when area==False is greater than the plot method.

fig,ax=pplt.subplot()
ax.sizelegend([50], area=False)
ax.plot(.2, .5, m='.', s=50)
ax.text(.2, .505, 's=50', ha='center')
ax.plot(.15, .5, m='.', ms=50)
ax.text(.15, .505, 'ms=50', ha='center')
ax.format(xlim=(.1, .3))

[cvanelteren]

I ended on that semantic legends now treat scatter-style size aliases as marker areas and convert them to Line2D marker diameters, while explicit markersize aliases remain diameters. This removes the extra area mode from categorical and entry legends, fixes the leaked scatter size alias, and keeps sizelegend area handling scoped to numeric levels.

[gepcel]

So, now the default value of area parameter is True for sizelegend and False for catlegend, right?

[cvanelteren]

No, the parameter is gone. You can directly select it as such:

  ax.catlegend(["A"], sizes=100)      # area keyword -> marker diameter 10
  ax.catlegend(["A"], s=100)          # area keyword -> marker diameter 10
  ax.catlegend(["A"], markersize=10)  # diameter keyword -> marker diameter 10
  ax.catlegend(["A"], ms=10)          # diameter keyword -> marker diameter 10

[gepcel]

Ok, I got the logic of catlegend. But I'm still a little confused. Matplotlib doesn't distingguite s/size and ms/markersize, right?

And I still suggest you change the diameter size of sizelegend to match matplotlib.

fig,ax=pplt.subplot()
ax.plot(.4, .65, m='.', ms=20)
ax.plot(.4, .45, m='.', s=20)         # s and ms of plot has the same size
ax.scatter(.6, .65, s=20, c='r')
ax.scatter(.6, .45, ms=20, c='r')  # s and ms of scatter has the same size
ax.sizelegend([20], loc='ll',area=False)  # I suggest size=20 match size of plot method
ax.format(xlim=(0,1), ylim=(0,1))

[cvanelteren]

LineCollections and PatchCollections internally don't use the same sizing.

The marker size in points**2 (typographic points are 1/72 in.). Default is rcParams['lines.markersize'] ** 2.

https://matplotlib.org/stable/api/_as_gen/matplotlib.pyplot.scatter.html

whereas plot will use points: https://matplotlib.org/stable/api/_as_gen/matplotlib.lines.Line2D.html#matplotlib.lines.Line2D.set_markersize