Why is smart_bounds() being deprecated?

[cristobaltapia]

I am getting the following warning when running my code that uses smart_bounds:

The set_smart_bounds function was deprecated in Matplotlib 3.2 and will be removed 
two minor releases later.

What should we use to get the same effect in the future then?

[jklymak]

What effect are you trying to achieve. This was relatively robustly discussed in the PR. You can look up the issue using the search function

[cristobaltapia]

I read the PR, but was not clear to me why the functionality was not deemed important. What I want is for the axis to span only the region where the datapoints are.

[jklymak]

ax.margins(0) does that, I think?

[cristobaltapia]

No, that is not what I meant. What I mean is e.g. if I set the limits of the y-axis to span [0-100] but my data spans [20-100], then the left spine would by drawn starting from 20 up to 100. But there will be an empty space from 0 to 20 (no line would be drawn).

[jklymak]

spine.set_bounds(x.min(), x.max())? https://matplotlib.org/3.1.1/api/spines_api.html#matplotlib.spines.Spine.set_bounds

I think the feature was deemed a bit strange to merit the amount of code and maintenance involved. But if you can make a strong case for it being useful folks may change their minds. (ping @efiring)

[tacaswell]

xref #11004

[cristobaltapia]

Well, I don't have strong sentiments attached to the function (although I have used it a lot in the past couple of years), but I think the functionality is nice. I read in Tufte's book [1] about the concept of "double-functioning labels", where the idea is to mark the range of your data with the spine. In the details it is not exactly as what set_smart_bounds() does, but quite similar. However, I guess that using spine.set_bounds(x.min(), x.max()) is also a viable solution.

[1] Tufte, Edward R. (1986). "The Visual Display of Quantitative Information", Graphics Press, Cheshire, CT, USA

[jklymak]

Sure, but why wouldn't you just set the xlim to [20, 100] and not have the blank space?

[cristobaltapia]

In the case where it is relevant to have the zero as well. If I am showing some ratios it helps to have the full range from zero, so that the magnitude is correctly interpreted.

[jklymak]

How does the viewer know zero is there somewhere? Maybe you could show an example?

[tacaswell]

One of the hardest things for us to know is what parts of the library are actually used. It is good to know this feature actually has users :)

I think the question is how are you using it and how onerous the alternatives are?

[cristobaltapia]

Here are some examples:

The first one is what I described above, where I have some ratios and it is important to know where zero is (so that large variations actually mean large variations and not just that we are in a different scale). I think is nice sometimes to just mark the range this way.

The other way I use it is in normal plots, where I just like the effect of the spines to be separated, like this:

Maybe in the first example (as pointed out by @jklymak) it would be better to also mark the zero in the y-axis. But those are my main uses for the function.

[jklymak]

Seems those would both work w/ set_bounds?

[cristobaltapia]

Yes, it would. The set_smart_bounds() is (as I see it) only a shortcut, especially useful when more than one data set is plotted (then the maximum and minimum of all the plotted data has to be considered).

[efiring]

Given that set_bounds seems adequate for the use cases, and has the advantage of being clear and explicit, I think this can be closed.