Wrap legend

[liamtoney]

Description of proposed changes

A wrapper for the GMT 6 legend command with support for specifying "handles" and "labels" as arguments, as in MATLAB (legend docs) and matplotlib (legend docs). The plotting commands (i.e., gmt.Figure.plot(), for now) are modified to return "handles" which are provided as inputs to gmt.Figure.legend(). This addresses part of #214 and #231.

Fixes#260

Reminders

• Run make format and make check to make sure the code follows the style guide.
• Add tests for new features or tests that would have caught the bug that you're fixing.
• Add new public functions/methods/classes to doc/api/index.rst.
• Write detailed docstrings for all functions/methods.
• If adding new functionality, add an example to docstrings or tutorials.

[weiji14]

Hi @liamtoney, just checking in on how you're doing. If you're stuck with finding a specfile, maybe try to pull in the @Table_5_11.txt dataset. You can look at GenericMappingTools/gmt#1749 (comment) for some inspiration (though the auto-legend won't work yet).

I also see that pylint is complaining a lot so I've suggested a few changes. Run make lint to see what the other problems are, and try your best to resolve those. If it's too hard, we can try to resolve them for you or teach you how to set an ignore flag (if it proves to be impossible).

[seisman]

The auto-legend feature has been merged into GMT 6.0 branch in GenericMappingTools/gmt#1749. Now we can use -l option in the plot module, which enables legends automatically. Below is a simple script showing how we can get auto-legends even without calling the legend module explicitly.

gmt begin plot png
gmt plot -R0/7/3/7 -Jx1i -B @Table_5_11.txt -Sc0.15i -Glightgreen -Wfaint -lApples
gmt plot @Table_5_11.txt -W1.5p,gray -l"My Lines"
gmt plot @Table_5_11.txt -St0.15i -Gorange -lOranges
gmt end show

This is what we will get:

With the auto-legend feature merged, it's almost straightforward to wrap the legend module in pygmt, and we can write a Python script like:

import pygmt
fig = pygmt.Figure()
fig.basemap(projection="x1i", region="0/7/3/7", frame=True)
fig.plot("@Table_5_11.txt", style="c0.15i", color="lightgreen", pen="faint", label="Apples")
fig.plot("@Table_5_11.txt", pen="1.5p,gray", label="My lines")
fig.plot("@Table_5_11.txt", style="t0.15i", color="orange", label="Oranges")
fig.show()

The only problem is, GMT 6.0.0 is not official released yet. It shouldn't take too long. We plan to release GMT 6.0.0 when the GMT6 paper is officially published.

[liamtoney]

Great. This should simplify the wrapper a lot. From a dev perspective, to build and test this wrapper do I now need to clone the GMT repo and build from source? Or are there plans for another conda release?

[weiji14]

We're still using 6.0.0rc4 for pygmt development, and that's what our code is tested against. I think we should wrap this up and refactor the code after 6.0.0 stable is released. The implementation doesn't have to be perfect, having something is better than nothing!

[seisman]

FYI, the GMT team is planing to release 6.0.0rc5 (with the auto-legend feature included) next week, and release the final 6.0.0 on Nov. 1.

[weiji14]

Right, it's not quite as easy as it seems... The auto-legend only gets triggered when we run gmt end, so the following won't work:

importpygmtfig=pygmt.Figure()
fig.basemap(projection="x1i", region="0/7/3/7", frame=True)
fig.plot(data="@Table_5_11.txt", style="c0.15i", color="lightgreen", pen="faint", l="Apples")
fig.plot(data="@Table_5_11.txt", pen="1.5p,gray", l="My_lines")
fig.plot(data="@Table_5_11.txt", style="t0.15i", color="orange", l="Oranges")
fig.show()

Produces:

I'm assuming though that calling legend would trigger the legend being plotted, is that correct @seisman？ I can start on some side work to wrap a "label" alias for text (and other functions). Plus we need to find a way to handle blank spaces so that label="My lines" works instead of having label="My_lines".

[seisman]

I think you're right. With pygmt, we have to call legend to plot the hidden legend specfile.

[liamtoney]

So could the code have an empty legend call after plotting, e.g.

importpygmtfig=pygmt.Figure()
fig.basemap(projection="x1i", region="0/7/3/7", frame=True)
fig.plot(data="@Table_5_11.txt", style="c0.15i", color="lightgreen", pen="faint", l="Apples")
fig.plot(data="@Table_5_11.txt", pen="1.5p,gray", l="My_lines")
fig.plot(data="@Table_5_11.txt", style="t0.15i", color="orange", l="Oranges")
fig.legend() # Empty legend callfig.show()

where fig.legend() when called with no arguments somehow accesses the hidden specfile if available?

[weiji14]

Hmm, I just tried using an empty fig.legend() on your branch (after making sure spec=None is allowed in the base_plotting.py's legend function) and it's complaining that the -D (a.k.a position) argument is needed. So again, not very automatic but at least we can get rid of the regex bits 🤷‍♂️

[weiji14]

Ah wait, I just noticed a new -T argument mentioned in the modern mode auto-legend note that outputs a specfile. We could perhaps set that to True if no specfile is given (i.e just a blank legend call) and use the auto-generated specfile to auto-generate the legend, what do you think?

Edit: Argh! Even with -T, we still need to use -D/position! Ok, just ignore this idea.

[liamtoney]

@weiji14, @seisman, @leouieda — what do you think about having a default argument for position? Say, JTR+jTR to by default place the legend in the top-right corner, inside. That way a user could have more fluid experience by simply calling pygmt.legend().

[weiji14]

Oops, sorry, hit the close button by accident...

I was thinking the same thing. The thing is, GMT does have a default position setting that gets triggered by calling end, and we would be 'reinventing the wheel' a bit here.

[weiji14]

Right, just a couple of minor edits to the test file (you're almost there!). Plus you'll need to add a line to the docs/api/index.rst file so that your shiny legend shows up in the docs! Do it like so:

 Figure.grdimage
Figure.legend
Figure.logo

Once you're done, remove the "WIP" from the title and it should be good to go 🚀. Let's open up a separate pull request to get that default position feature into PyGMT since this one is already getting way too long.

[liamtoney]

@weiji14 I've made those changes. My commit history is a little messy (sorry) — does anything need to be done about that prior to merging?

[weiji14]

Hi @liamtoney, just one last change! And no you don't have to worry about the commit history since we'll be squash merging (i.e. it'll only show up as a single commit on the master branch).

[liamtoney]

Thanks for walking me through this, @weiji14!

[weiji14]

Phew, good work @liamtoney. Thanks for having the stamina to see this one through! Feel free to open up another pull request to get the auto-position done (and we can debate around the implementation 😆). Also it'd be great to have a gallery example for legend on the website.

[seisman]

grdcontour doesn't support -l option now. Currently only plot supports auto-label.

[weiji14]

Resolving in #357.

[weiji14]

Oops, that's my bad 🤦‍♂️. I'll fix that.