include validation function that makes sense for individual-based simulators

[grahamgower]

We currently allow things that are permissible in a coalescent framework, but makes no sense for individual-based and/or discrete-time simulators. E.g.

• 0 < N < 1, and
• epoch and migration time spans less than 1 generation in length (either directly, or indirectly when using time_units!="generations").

One way for an individual-based simulator to check this, is to convert the relevant numbers to integers, then rebuild the graph. This will be (or should be) a fairly common operation, so it seems reasonable that we should provide such functionality as part of demes.

What if we had something like this?

class Graph:

    ...

    def frob_time_and_size(self, func):
        """
        Return a copy of the graph with times and sizes modified by func.
        """
        b = Builder.fromdict(self.asdict())

        for deme in b.get("demes", []):
            deme["start_time"] = func(deme["start_time"])
            for epoch in deme.get("epochs", []):
                epoch["start_time"] = func(epoch["start_time"])
                epoch["end_time"] = func(epoch["end_time"])
                epoch["start_size"] = func(epoch["start_size"])
                epoch["end_size"] = func(epoch["end_size"])

        for migration in b.get("migrations", []):
            migration["start_time"] = func(migration["start_time"])
            migration["end_time"] = func(migration["end_time"])

        for pulse in b.get("pulse", []):
            pulse["time"] = func(pulse["time"])

        return b.resolve()

And then a simulator would call graph.frob_time_and_size(int). Or to scale the simulation by a factor Q: graph.frob_time_and_size(lambda x: int(x / Q)).

[jeromekelleher]

Seems reasonable - I expect @molpopgen will have thoughts?

[grahamgower]

On second though, it should probably be frob_time_and_size(self, time_func, size_func), so that Graph.in_generations() can be implemented using it.

[molpopgen]

I'm starting to dust the cobwebs off here and will be thinking about this soon. @apragsdale has some prototype code that'll be the basis for the next fwdpy11 release. I'm sure we will come across what's necessary to check while working on that.

[molpopgen]

@apragsdale -- let's add a test of this when working on the fwdpy11 integration. We can just make a model with some small time intervals.