plot vs scatter has two aspects:

• The technical one: Are all markers the same?
• The semantic one: What do we assume the data does represent:
  plot implies some sort of ordering, which is reflected by the default of drawing a line through the sequence of points.
  • scatter assumes the points are uncorrelated and (x, y) are both coordinates. This is also reflected by the fact that scatter accepts 2D arrays, e.g. from meshgrid.

Therefore the note in scatter

The plot function will be faster for scatterplots where markers don't vary in size or color.

does not sufficiently cover the topic, e.g.

x = y = np.linspace(0, 1, 11)
X, Y = np.meshgrid(x, y)

_, (ax1, ax2) = plt.subplots(2, 1)
ax1.plot(X, Y)
ax2.scatter(X, Y)

I have no clear idea how to accomodate both aspects, but we should not just blindly follow the technical argument.

There's actually another reason to prefer plot() to scatter() when variable markers are not needed (but I didn't want to get into the details here): scatter autoscaling on log scale is poor (#3059, #5897, #10782, #11898), and one can easily hit that e.g. when changing the scales interactively. It is possible that #13642 will improve the situation, but it's not merged yet.
Also, I think in many cases the default marker size from plot() is just preferable to the one from scatter(); compare e.g. the scatter_hist.py example before (using scatter()):

and after the PR (using plot()):

(Obviously, one can change the marker size in either function call, so that's not a very strong argument.)

Marker size is a style choice, it really depends on your application. You may be right with log scaling.

However, in the end, this supports my point: There's more difference between plot and scatter than "plot is a faster substitute for scatter with constant color and marker size".

I'm open which one to use under which circumstances, but that needs clear documentation and instructions. Without that e.g. using plot instead of scatter in an example titled "Scatter plot with histograms" is just plainly confusing.

I can just revert the two "scatter hist" examples for now? None of the other examples mention the term "scatter", and I have no doubt there's a bunch of other places in the examples where we use plot(..., ".") to plot a point cloud.

The "Scatter hist" are the two most obvious scatter-like ones. The only clear case in which I really prefer plot is the "Violin plot customization". The other examples semantically plot scattered points. So I wouldn't change them.

I won't give a positive review until we've documented the aspects discussed above #14174 (comment). But if you at least leave out "Scatter hist", I'm not blocking this if others choose it's ok.

Reverted the scatter hist examples.

Since this Pull Request has not been updated in 60 days, it has been marked "inactive." This does not mean that it will be closed, though it may be moved to a "Draft" state. This helps maintainers prioritize their reviewing efforts. You can pick the PR back up anytime - please ping us if you need a review or guidance to move the PR forward! If you do not plan on continuing the work, please let us know so that we can either find someone to take the PR over, or close it.

Let's not bother.