Constraining multiple nodes to be at the same time?

[hyanwong]

If we have full diploid sequences from an undated historical sample (unusual, I know), then we would want to constrain both nodes to be at the same time.

Similarly, if we have a full ARG in which a single recombination event is represented by 2 recombination nodes (as in msprime), then we would want to keep the nodes at the same time too. See kitchensjn/tskit_arg_visualizer#41 (comment)

I can't think of a nice API for this, though! Perhaps if multiple nodes are in the same individual we should treat them as having identical times?

[nspope]

I think we want to be able to pass in custom priors for particular nodes. Typically, this would be shape/rate of a gamma -- but it could also be the ID of another node, in which case we would make both nodes share the same posterior.

Alternatively, we could have a separate argument, node_equivalence, which would be a list of (non-overlapping) sets of nodes.

In effect, messages passed to either node would be aggregated into the same posterior. Technically this is pretty easy to do, but definitely would involve some rejiggering of the code.

[hyanwong]

Ah, the ID of another node sounds like a plan (and might help with the prior too, if we ever allocate different ones to different nodes.

I'm not sure this is a priority, but it does seem like something we might want to do eventually, right?

[nspope]

I think so? For instance if a tree sequence is inferred in tandem with a known pedigree?