BlockSampler validation of a graph with self edges

[rsheth-dwave]

When attempting to instantiate a BlockSampler on a fully-connected GRBM w/ trivial coloring, lambda x: x, BlockSampler._valid_colouring() returns False. It is failing on self edges. Should it ignore these for validating the coloring?

[kevinchern]

I think the fix should be at the GRBM level. Self-edges should be disallowed at construction since $s^2=1$

[passionworkeer]

I’d handle this before coloring validation. If self-edges are mathematically redundant for the GRBM case because s^2 = 1, the constructor (or a graph normalization step) should either strip the diagonal or raise a clear ValueError when diagonal entries are present. Then _valid_colouring() can stay focused on off-diagonal conflicts only. That avoids a false negative in coloring while also making the model semantics explicit at the API boundary.

[VolodyaCO]

@rsheth-dwave could you elaborate on the use-case for having a GRBM with self edges?

EDIT: are you using the GRBM for something that is not an interacting system of two-level systems?

[rsheth-dwave]

@rsheth-dwave could you elaborate on the use-case for having a GRBM with self edges?

EDIT: are you using the GRBM for something that is not an interacting system of two-level systems?

@VolodyaCO There isn't a use case w/ self edges I was considering. I was wondering whether the returned validation error may belong one level above BlockSampler.

[VolodyaCO]

Ok, so in that case I agree with @passionworkeer that there should be a check in the GRBM constructor so that no self edges are allowed.