Currently, (advanced) indexing, slicing, and masking of a Normal random variable with a LinearOperator mean and/or cov simply calls .todense() on the parameters and performs all operations on the matrices.
At least for simple indexing and slicing this can be sped up drastically by using broadcasted matrix vector products.
We should also investigate whether efficient advanced indexing and masking is also possible.
marvinpfoertner
commented
4 years ago
Currently, (advanced) indexing, slicing, and masking of a
Normalrandom variable with aLinearOperatormean and/or cov simply calls.todense()on the parameters and performs all operations on the matrices.At least for simple indexing and slicing this can be sped up drastically by using broadcasted matrix vector products. We should also investigate whether efficient advanced indexing and masking is also possible.