Crosscat hyperprior grid on variance parameter is broader than it needs to be about crosscat

[riastradh-probcomp]

Sounds reasonable to me. What about the lower bound? .01*sum((x-x̄)^2) as we currently use seems pretty arbitrary to me -- surely there could be clusters with much smaller variance than that, very far away from one another.

[riastradh-probcomp]

You could specify 0 as a lower bound for log_linspace, and you would get a grid starting at the smallest positive normal floating-point number.

[riastradh-probcomp]

I suppose the smallest distance between any pair would be a better starting point, but that takes O(n^2) time to compute.

[alxempirical]

I guess you can do it in O(n*log(n)) by sorting, since closest pairs will be adjacent in the sorted list.

[riastradh-probcomp]

Right -- I was inexplicably thinking of >1-dimensional spaces. That would probably be a reasonable thing to do, then.

[alxempirical]

I think you can probably even do closest points in high-dimensional spaces with a KD-tree.

[alxempirical]

Oh, there's a wikipedia page about this exact problem.

[riastradh-probcomp]

Golly, my memory of computational geometry has rotted.

However, 0 may also nevertheless be a reasonable choice to start anyway -- for isolated outlying clusters we don't have a reasonable lower bound on their variance.

[alxempirical]

Yes, the current lower bound seems likely to cause a problem.