Should we use n by 1 or length n for the size of a vector, eg. [0, 0]?
There is some inconsistent issue with these in the mvn.
For example,
dist1 = MultivariateNormal(mu1=[0, 0], covariance_matrix=[[1, 0], [0, 1]])
dist2 = MultivariateNormal(mu2=[[0], [0]], covariance_matrix=[[1, 0], [0, 1]])
when we do the VariableCast for mu1 and mu2, they will be squeezed and turned into [0,0] and then a Variable. In other words, both mu will have size 2. However, if we draw sample from dist1 and dist2, which should be a vector and have the same size of mu, the size of the sample will be 1 by 2. It will be much clear in debug.
We need to solve this: either squeeze the sample or always ensure 1 by n or n by 1 for the size of vector.
Should we use n by 1 or length n for the size of a vector, eg.
[0, 0]?There is some inconsistent issue with these in the mvn.
For example,
when we do the VariableCast for mu1 and mu2, they will be squeezed and turned into [0,0] and then a Variable. In other words, both mu will have size 2. However, if we draw sample from dist1 and dist2, which should be a vector and have the same size of mu, the size of the sample will be 1 by 2. It will be much clear in debug.
We need to solve this: either squeeze the sample or always ensure 1 by n or n by 1 for the size of vector.