Document GRM estimation using mean imputation.

[timothymillar]

It's common practice to use mean imputation for missing data when calculating the VanRaden GRM. We should add an example of this to the docstring. The following example avoids needing to call compute on intermediate variables:

import numpy as np
import xarray as xr
import sgkit as sg

ds = sg.simulate_genotype_call_dataset(n_variant=6, n_sample=5, missing_pct=0.1, seed=0)
sg.display_genotypes(ds)
ds = sg.call_allele_frequencies(ds)
# mean imputation
ds["variant_allele_frequency"] = ds.call_allele_frequency.mean(dim="samples", skipna=True)
ds["call_allele_frequency_imputed"] = xr.where(
    ~np.isnan(ds.call_allele_frequency),  # where call allele frequency is not missing
    ds.call_allele_frequency,  # use the call allele frequency
    ds.variant_allele_frequency,  # else use the variant mean allele frequency
)
ds["call_dosage"] = ds["call_allele_frequency_imputed"][:,:,1] * 2  # multiply by ploidy
ds["ancestral_frequency"] = ds["variant_allele_frequency"][:,1]  # use mean frequency as ancestral frequency
ds = sg.genomic_relationship(ds, ancestral_frequency='ancestral_frequency')
ds.stat_genomic_relationship.values 

[jeromekelleher]

Looks good @timothymillar. I wonder if we should have a "Pedigree statistics" top level section with some narrative and examples though, where something like this could go? I was actually just looking for documentation this morning answering a question about computing inbreeding from simulations, and I think it would be good to have some top-level docs here to orient new users.