search

statnet / ergm

Fit, Simulate and Diagnose Exponential-Family Models for Networks
Other
98 stars 37 forks source link

fixedas(present, absent) constraint does not actually ensure edges are present or absent. #572

Closed krivit closed 1 month ago

krivit commented 4 months ago

The constraint simply takes the provided edgelist and fixes the corresponding dyads. It doesn't check if the edges in present are present, in absent are absent, and it doesn't modify the network such that they are. Passing an argument as present has the exact same effect as passing it as absent.

It was originally implemented by Kirk Li in response to a ticket by @CarterButts (https://github.com/statnet/ergm-private/issues/49), so the question is, what do we do with it?

krivit commented 4 months ago

One option would be to add another argument in front, e.g., fixed.dyads, which doesn't get checked, then if the user passes present or absent by name we check for presence and absence on initialisation and errors out if violated. In principle, we could come up with a mechanism to bring the LHS network into conformance with the constraint, but that requires developing a whole new API.

martinamorris commented 4 months ago

I guess I'm not sure what the use case is here. Was the goal to constrain some dyads to their initial values, and allow others to be toggled? Or was it supposed to fix some dyads' statuses to present/absent and allow others to be toggled.

krivit commented 4 months ago

The main use case is to fix (i.e., condition on) a specific set of dyads at their initial values. I don't know if anyone has relied on the present/absent functionality---in which case they would get incorrect results.

krivit commented 3 months ago

@CarterButts , @martinamorris , @drh20drh20 , @kecoli thoughts?

martinamorris commented 2 months ago

I don't have strong feelings about this -- but unless there's a strong use case made for the present/absent arguments, it seems easiest to deprecate them and just use this constraint for setting a passed edgelist to the observed values.

krivit commented 1 month ago

Since there isn't currently an API for a constraint to modify the network, the most it can do anyway is to raise an error. So, I guess the best I can do is