9 mixture distributions

[craig-parylo]

Closes #9

First draft to include @anyaferguson 's mixture distributions as an accompanyment to the point-range plots.

Settings

'Show ECDF' - unchecked (default) shows the probability density, checked shows the cumulative distribution.

'Show NEE range' - checked (default) enables a crossbar geom that overlays the results of the national elicitation exercise (similar to the point-range plots)

The process to generate mixture distributions takes a couple of seconds to calculate and a few more to render as plots. Not a lot of time in the scheme of things, but something I think would be noticeable to the user. In this PR I've arranged for the mixture distributions to be pre-rendered on the app start-up based on all available schemes to speed up the chart generation.

Note, for the future we may want to add in an invalidation timer to ensure the mixture distributions are regenerated on a routine basis - preferably in sync with the updates to the underlying data.

Known issues

🐛 Mixture distributions are fixed to include all schemes
Pre-calculated mixture distributions (see above) designed to improve UX by increasing responsiveness also means they aren't dynamic, so include the focal scheme and do not respect the comparator schemes selection.

🐛 ggplot2 warning
Displaying the NEE reference range results in the below warning message printed to the console:
Warning in ggplot2::geom_crossbar(data = stats::na.omit(focal_pointrange), : Ignoring unknown aesthetics: width

This is caused by referencing width when setting the aesthetics for the geom_crossbar() - which is my attempt at making the width (actually the height) of the crossbar dynamic to the maximum height of mixture distribution plot. Attempts to resolve include a) removing the width argument and b) setting a fixed width. Results show a) results in a very narrow crossbar to appears as a single line and mid-point is not visible, b) uneven widths relative to the mixture distribution plot across facets - the y-axis is variable scale depending on the maximum density of the distribution plot, so some facets with a more even distribution show a crossbar with large width and others with large peaks have a narrow crossbar width.

[matt-dray]

Cracking, many thanks. We talked about this earlier today. As mentioned, I'm approving it and we can iterate later on the issues you mentioned.

However, can you add one thing before you merge? Could you suppress the 'width' warning and add a comment in the code to say that the warning is known about?

I'll add a few techdebt issues to the repo in light of this:

• Revisit the composition of the mixture-distribution aggregate #83
• Fix warning about 'width' as an unknown aesthetic #82
• handle the case where no data is shown (e.g. if Leighton is selected as the only scheme to visualise), though this may come under Warn user with meaningful message when there's insufficient data (point-ranges, heatmaps) #30

I've created a more general issue (#80) about how best to describe plots as well, which we also chatted about.

Cheers.