start of plotting "help" module

[philippemiron]

• clean up codes
• wraps the import to matplotlib and cartopy to include an Optional dependencies warning
• validate that a projection is passed if using a geoaxes
• add docstring
• create examples
• tests

As of right now, the goals are:

• implement How to do a global spaghetti plot with trajectories crossing the dateline or edges on longitude range? #200
• create a function to return a LineCollection to avoid using apply_ragged() with matplotlib functions (to fix apply_ragged with cartopy #227).

[selipot]

Great start! maybe call the module plothelp or helper instead of help which is a bit too general?

Maybe the two goals could be combined? i.e. the function that returns the LineCollection returns extra lines when crossing the dateline to accommodate? How do we make sure that the assigned colors of the lines for the drifters that cross the dateline are the same?

[milancurcic]

What do you think about utility? For the function that inserts NaN at dateline, I suggest making a function that inserts NaNs between segments, and let the user segment the dateline. That could fit in the analysis module. I don't quite see the need for a separate module here. I know @selipot suggested it and I didn't say anything, but I don't see it as compelling.

[milancurcic]

and let the user segment the dateline

We can then have a separate function that segments the dateline specifically, so the desired functionality here would be by chaining the two functions.

[philippemiron]

I have no preference about the name of the module and/or the location of those functions.

I was going to make the first function to insert nan general enough to deal with any periodic domain, expanding on this:

nan_indices = np.cumsum(cd.analysis.segment(lon, 180, rowsize=cd.analysis.segment(lon, -180)))[:-1]

I can switch it to utility.py for now and we see where we put the functions once ready.

[philippemiron]

Great start! maybe call the module plothelp or helper instead of help which is a bit too general?

Maybe the two goals could be combined? i.e. the function that returns the LineCollection returns extra lines when crossing the dateline to accommodate? How do we make sure that the assigned colors of the lines for the drifters that cross the dateline are the same?

If we include nans at the right place, we don't need to split trajectories into 2 segments on either side of the dateline and duplicates colors to properly plot them. And at the same time, it would limit to a single task the function that generates the LineCollection.

[philippemiron]

Not sure exactly how to proceed now, segment only supports 1-d array... 😭. And I also need a way to update rowsize after adding the nan.

[philippemiron]

@selipot want to test? I guess I might have some time next week during Thanksgiving to play with this.

[philippemiron]

I'm abandoning the idea of rebasing this branch. Many files have been renamed, deleted and functions moved around, making this very annoying. I might just reopen another PR once we are closer to merging this.

[selipot]

I'm abandoning the idea of rebasing this branch. Many files have been renamed, deleted and functions moved around, making this very annoying. I might just reopen another PR once we are closer to merging this.

Sorry about that, that happens when a PR is open for too long for a version 0.x software I guess. Do you want me to test before you open another PR or after?

[philippemiron]

No worries, the easiest would be just to copy the function.

[philippemiron]

One question, how can I test something that requires optional dependencies... 🤔 I guess we would need to have a different environment for the test, which would include all the mandatory and optional packages.

[milancurcic]

How about installing all dependencies (core and optional) for testing?

[selipot]

Testing your function. I think there is an issue with your first test lines 132-150. It raises a valueError if ax is a simple plt.Axes which it should as it is written. As an example this does not work:

fig = plt.figure()
ax = fig.add_subplot(1, 1, 1)
plot_ragged(ax, ds.lon, ds.lat, ds.rowsize, colors=np.arange(len(ds.rowsize)));

[selipot]

The examples need editing which I can do once you submit a new PR.

[philippemiron]

Did you try the latest version? You can use this branch, I fix it yesterday.

[selipot]

I copied the code for the function from the PR branch as you instructed. Again, lines 132-150 don't make sense to me or rather the error I get makes sense. Why would you test for ax to have the attribute coastline if it is only plt.Axes?

if isinstance(ax, plt.Axes) and hasattr(
         ax, "coastlines"
     )
...

[philippemiron]

If it has attributes coastline, it is a geoAxe and we have to also try to load cartopy. This way, the function can be used with matplotlib only.

The else condition should be one (or two) level(s) further aligned with the if isinstance(geoAxe). I'm on the plane but can fix in ~2h.

[selipot]

Yes but the exception to your if statement is an error so that you get an error still if the axis is simply plt.Axes. The function works for me if I comment that if statement.

[philippemiron]

@selipot fix now.

[selipot]

What's next with this @philippemiron ?

[philippemiron]

Did you have chance to give it a try?

[selipot]

Yes! Works for me. But did not test with cartopy.

[milancurcic]

This looks good to go to me. I installed it, plotted some stuff, looks good. Tests pass after installing matplotlib and cartopy.

utility as a module name is quite generic. Should we call this plotting?

[milancurcic]

**kwargs are below. LGTM.

[milancurcic]

Is there a mechanism to skip these tests if optional dependencies are not installed?

[philippemiron]

True, I assumed that it was installed in the ci/cd, but I see that this wouldn't work if you don't have those packages locally.

[philippemiron]

Can you try again with the latest modification?

[milancurcic]

I forgot one important thing: we should add a note to the installation instructions (in README.md and docs/install.rst) about installing optional dependencies.

[milancurcic]

I think we discussed and established that matplotlib-core is already implicitly installed by depending on xarray, correct? In that case, I think only cartopy should be an optional dependency.

[philippemiron]

If I installed the environment I don't get matplotlib.

[philippemiron]

I think we discussed and established that matplotlib-core is already implicitly installed by depending on xarray, correct? In that case, I think only cartopy should be an optional dependency.

do we really need this? If someone tries to import the module it already say:

raise ImportError("missing optional dependency 'matplotlib'")

[milancurcic]

My bad... I got confused about matplotlib-core. Not installed in my venv either.

[milancurcic]

@selipot please check on your end if all good, and merge when ready.

[selipot]

just small editing of the docstring examples needed.