QC for OutOfAfrica_4J17

[jeffspence]

QC for #733 (@grahamgower)

This is my first time doing QC -- so let me know if I'm doing anything wrong. I was able to (blind) replicate the model. The only issues I ran into were:

1. The times in generations as currently implemented have all been cast to integers. e.g., in the notation of Table 4 here T_{CH-JP} is 9kya. The generation time is 29y so I originally implemented this as 9e3/29 but that caused the tests to fail. I could get the tests to pass if I changed this to int(9e3/29). The models are inferred by moments which assumes discrete (hence integer) generations so I guess that makes sense, but since they only report to the precision of 9kya, it's not clear whether it makes sense to convert to an integer or not.
2. I initially used the newer msprime API to specify the population splits (add_population_split) but that caused one of the populations to become "inactive" which caused the tests to fail. I could get around this by switching to the old style of specifying the equivalent population splits as mass migrations, plus setting migration rates to zero. I guess this is maybe a personal preference kind of thing, so I'm happy to go either way.

otherwise my initial implementation matched the current implementation.

[jeffspence]

I should also say that my first commit on this branch is the model as I originally implemented it, and the second commit is the tweaks I mentioned above (switching generation times to integers, and switching to add_mass_migration)

[codecov]

Codecov Report

Merging #1246 (0294f0e) into main (4d01954) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main    #1246   +/-   ##
=======================================
  Coverage   99.56%   99.57%           
=======================================
  Files         103      103           
  Lines        3248     3296   +48     
  Branches      415      415           
=======================================
+ Hits         3234     3282   +48     
  Misses          6        6           
  Partials        8        8           

Impacted Files	Coverage Δ
stdpopsim/qc/HomSap.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4d01954...0294f0e. Read the comment docs.

[grahamgower]

LGTM, thanks @jeffspence.

[grahamgower]

QC for #733 (@grahamgower)

This is my first time doing QC -- so let me know if I'm doing anything wrong. I was able to (blind) replicate the model. The only issues I ran into were:

1. The times in generations as currently implemented have all been cast to integers. e.g., in the notation of Table 4 here T_{CH-JP} is 9kya. The generation time is 29y so I originally implemented this as 9e3/29 but that caused the tests to fail. I could get the tests to pass if I changed this to int(9e3/29). The models are inferred by moments which assumes discrete (hence integer) generations so I guess that makes sense, but since they only report to the precision of 9kya, it's not clear whether it makes sense to convert to an integer or not.

Conversion to intergers is fine here. It makes sense in the context, and there will be no practical difference in simulation output as a result.

2. I initially used the newer msprime API to specify the population splits (add_population_split) but that caused one of the populations to become "inactive" which caused the tests to fail. I could get around this by switching to the old style of specifying the equivalent population splits as mass migrations, plus setting migration rates to zero. I guess this is maybe a personal preference kind of thing, so I'm happy to go either way.

This can be dealt with by using the population_id_map parameter to DemographicModel. But its undocumented, so don't worry about it.

otherwise my initial implementation matched the current implementation.

Great! I'm going to merge this without squashing, because your first model version could be useful. Thanks!