Arg diagram

[jeromekelleher]

First pass for #85

Here's what it looks like:

There's a few problems:

• The ARG illustration is using tikz, while the tskit output is svg. We'll never get the fonts to match, which would be horrible. Maybe someone could do a version of the ARG drawing in inkscape, which would hopefully use the right fonts?
• The tskit ARG isn't technically the same as msprime's output, as msprime would put in two nodes for the recombination, not one. I originally did it this way to keep things simple for people in a talk, but it's perhaps not such a good idea here.
• I had to use Chromium to convert the pdf to get the formatting to work, which meant I had to specify the sizes in the SVG. Will need to be looked at properly, probably.
• It's quite busy, and I'm not sure that (D) adds much now.

WRT to (D), if we go with this illustration perhaps we could just drop the plot and instead give a concrete example in the text? Something like

The extra ARG information quickly comes to dominate in the output tree sequences, since most common ancestor events don't result in a coalescence and most recombination events are not seen in the output trees. For example, for a simulation of 100 samples with a sequence length of 100 megabases and recombination rate of 10^{-8}, 9x% of the rows in the node table are concerned with recording these events, and the overall tree sequence is XXX times larger than its simplified equivalent (which is precisely what is output when the record_full_arg options is False)."

[JereKoskela]

It doesn't strike me as too busy, but your paragraph looks like a good substitute for (D) if I'm in the minority.

I have no experience of inkscape, but would it not be equally good to draw pictures equivalent to the tskit output in tikz? It looks like it should be straightforward since all the branch length y-coordinates are already present in (A).

[grahamgower]

I have no experience of inkscape, but would it not be equally good to draw pictures equivalent to the tskit output in tikz? It looks like it should be straightforward since all the branch length y-coordinates are already present in (A).

There's a tskit/tikz pr open here.
tskit-dev/tskit#798

[jeromekelleher]

I think it'll be less work overall if we use the SVG output, and convert the ARG into SVG - the SVG output is a lot more mature, and especially if we're showing other tree sequences, we'll want to keep things uniform.

OK, I'm going to commit this much for now, we can work on from there.