I’ve fit a model with SVI using a AutoNormal
guide, which runs fine, but when I want inference on the results, via quantiles
:
estimates = guide.quantiles([0.025, 0.5, 0.975])
I get a runtime error:
RuntimeError: The size of tensor a (441) must match the size of tensor b (3) at non-singleton dimension 0
Curiously, this works fine with AutoLowRankMultivariateNormal
and AutoMultivariateNormal
. I would have expected it to work here to. Am I overlooking anything?