You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
One of the changes that is being made with PR #538 is some performance improvements to the construction of dense elementary error generator matrices. The construction of these matrices is an important primitive in many of the error generator related functions and classes throughout pyGSTi. See this commit for the relevant changes, 7733720, but the thing being leveraged here is the fact that the construction is always performed initially in the standard basis. This means that the explicit matrix multiplications that were being performed can be replaced by a combination of vector outer products instead for a performance boost. There are also some specializations implemented now for the pauli case that let us skip some extra operations.
One bit of extra structure that isn't being exploited right now is the fact that the paulis are generalized permutation matrices, so the outer products we're doing are between single-element vectors. We should in principle be able to use this to avoid even any explicit outer products and build the matrices directly using just indexing/assignment operations which likely will be even faster.
The text was updated successfully, but these errors were encountered:
One of the changes that is being made with PR #538 is some performance improvements to the construction of dense elementary error generator matrices. The construction of these matrices is an important primitive in many of the error generator related functions and classes throughout pyGSTi. See this commit for the relevant changes, 7733720, but the thing being leveraged here is the fact that the construction is always performed initially in the standard basis. This means that the explicit matrix multiplications that were being performed can be replaced by a combination of vector outer products instead for a performance boost. There are also some specializations implemented now for the pauli case that let us skip some extra operations.
One bit of extra structure that isn't being exploited right now is the fact that the paulis are generalized permutation matrices, so the outer products we're doing are between single-element vectors. We should in principle be able to use this to avoid even any explicit outer products and build the matrices directly using just indexing/assignment operations which likely will be even faster.
The text was updated successfully, but these errors were encountered: