Jerome Kelleher

We had a discussion about this earlier and @brieuclehmann is going to run it through [memory-profiler](https://pypi.org/project/memory-profiler/). Assuming we're on the right track memory reduction wise, the plan is to add...

> Is memory-profiler measuring actual process memory, Yes, it takes periodic snapshots using OS routines, so it's the "real" process footprint.

Pasting in the profiles for ease: ``` ilename: test_time.py Line # Mem usage Increment Occurences Line Contents ============================================================ 9 65.5 MiB 65.5 MiB 1 @profile(stream=fp) 10 def genetic_relatedness_matrix(ts, sample_sets, mode):...

There isn't a huge difference either way here - maybe run this on a later example? Also do just one thing on the line where we call ``genetic_relatedness``, the assignment...

> For context, when n_samples = 2000 in the above examples, we have num_nodes = 6068. Ah, I see what's happening here - your tree sequence is quite short, so...

Still surprisingly little...

Ah - I don't think this is computing the right value @brieuclehmann. Tests are failing because we're getting different numbers. I'll think about how to do this again.

Agreed. As discussed somewhere in #261, I think the ``mask`` argument should be a 2D numpy array with shape (N, 2) so that we would mask out two regions with...

As discussed, we're not sure when these are needed though. I'm provisionally adding them to a new 0.2.3 milestone, but we can promote to 0.2.2 if we want.

Yes - but it's a can of worms. I wouldn't take it on unless you have a spare month!