Add P3 examples

[RobertPincus]

Adds a few hopefully interesting examples focusing on measurements from the P-3. Uses the colorcet, seaborn and especially cartopy Python modules. Cartopy has many dependencies so modules are now installed via conda for the Continuous Integration. Adding cartopy closes #50 and may be relevant to #11.

Probably best to squash-and-merge as no one needs to see my stumbling around.

[RobertPincus]

To install Python dependencies with conda: conda env create -f environment.yml will create an virtual environment how_to_eurec4a. Using source activate how_to_eurec4a or conda activate how_to_eurec4a will make the installed modules available.

[RobertPincus]

... and probably we need to update gitlab-ci.yml to install dependencies and invoke the virtual environment with conda rather than `pip.

[RobertPincus]

@tmieslinger In case you have the bandwidth to look over these changes (@d70-t being quite busy it seems), I would prefer not to merge them until someone else had a look at them.

Note that the cell timeout is currently set to 150 seconds; this sometimes is too short for the P-3 flight track cells.

Also - we're now using conda to manage Python dependencies in the Github CI but I don't know how to make this change on the Gitlab which builds the book for distribution.

[d70-t]

Sorry for not responding yet. I generally like the changes a lot, but had to do something else the last two days. I've got a few comments and will likely post a partial review soon.

[RobertPincus]

Of course, @d70-t, and no worries. Not everything should fall on your shoulders.

I quite liked your cleanup ideas, I have to say (#47)

[RobertPincus]

@d70-t Cool, thanks, I'll take a look. The book chapters are just excerpts from the notebook that generates figures for the paper, which is why there's overlap...

[RobertPincus]

@d70-t How should we proceed with cartopy and pip? I understand the issue to be that pip doesn't install all the needed dependencies, and that we expect people to have done this on their own.

[d70-t]

Initially I thought the best solution might be to include cartopy in requirements.txt as this would require one step less to install manually (users would only have to install cartopys dependencies, not cartopy itself manually).

But now I am thinking, maybe it is actually better not to specify it in requirements.txt, because that way, the installation won't fail if the dependencies of cartopy are not there. In stead only those notebooks which actually use cartopy will fail. If someone wants to run / edit the book without having cartopy installed, this would then still be possible (one would have to ignore some build errors though, but that's probably ok?).

If a user who doesn't use conda is more in the cartopy-business, I think it is not significantly more complex to install cartopy's dependencies and cartopy itself.

I'd suggest to update the running locally document (which should soon turn into a few more chapters describing how to contribute, prepare datasets etc...) such that it contains at least a description on how to set the book up using conda. I'd also keep the description on pip and add a short note that one should try to get cartopy running as well. (This might be as simple as apt install python3-cartopy or something similar using brew, but I'd not go into further details. For the two alternatives, tabbed blocks or panels might be handy. I'd be happy if you could write something about conda, because I really don't know that thing. If you don't want to mess too much on pip here, I'd suggest to make this PR such that it runs smoothly on conda and is properly documented. I'll then create a new PR on top of this which tries to get a new pip based thing running again as well.

[RobertPincus]

@d70-t I added material to the chapter on running locally to describe the use of conda. You can let me know if it's not clear. I don't mention, but maybe I should, that conda is much slower than pip, partly but not exclusively because it's managing cartopy and its dependencies.

[d70-t]

As @tmieslinger noted, it seems like I've forgotten to include the color_of_day function in the review:

def color_of_day(day):
    return plt.cm.viridis(norm(day), alpha=0.9)

[RobertPincus]

@d70-t @tmieslinger Thanks for your very helpful comments. I believe they are all incorporated including a little bit of extra explanation about how to run locally. I've explained that users need to manage the installation of cartopy themselves.

I have not updated the gitlab-ci to work with conda so the book won't publish automatically until/unless you adopt the Github publishing @d70-t is working on with #53 .

As of now I won't plan to make further changes until the publishing issue is resolved, but it would be nice to have this folded in sooner than later.

[RobertPincus]

I would suggest that you add yourselves as co-authors to the commit.

[RobertPincus]

@d70-t The speedup was an excellent suggestion, thanks a lot, and I think all the changes are in place. The CI is failing in the HAMP chapter for reasons I don't understand, but we can re-run the CI before merging once the domain is in place.

[d70-t]

The HAMP files have been moved. That should be fixed in eurec4a/eurec4a-intake#67, but maybe I'll have to add another quick fix to the book as the version update could lead to a minor issue in some of the chapters... I'll sort it out.

[RobertPincus]

@d70-t @tmieslinger I brought the P-3 pull request up to date with the excellent changes on master. I hesitate to squash-and-merge because I don't understand what will happen to the Gitlab CI, which uses pip without cartopy, while the Github CI and publication uses conda