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
The key idea is to extend Pyomo's representation system to enable the expression of matrix/vector representations. While this does not improve the efficiency of Pyomo's core expression system, it would offer users an alternative that is likely more performant. Generally, Matrix/Vector expressions would be smaller, and hence the operations needed to canonicalize them would be a smaller burden when interfacing with solvers.
Note that this we have a preliminary Matrix representation in Pyomo kernel. However, a key element of this approach would be to define a Vector representation for parameters and variables. Or to define a convention for using Pyomo's existing Variable/Parameter objects in that manner. But for the hackathon our goal should be to demonstrate the potential performance win.
NOTE: I think it would make sense to include the use of numpy matrix/vector data and scipy.sparse matrix/vector data in our performance hacking. Many users will naturally gravitate to the use of these data formats, and thus I think it would be a mistake to limit the scope of this activity to custom matrix/vector objects.
The key idea is to extend Pyomo's representation system to enable the expression of matrix/vector representations. While this does not improve the efficiency of Pyomo's core expression system, it would offer users an alternative that is likely more performant. Generally, Matrix/Vector expressions would be smaller, and hence the operations needed to canonicalize them would be a smaller burden when interfacing with solvers.
Note that this we have a preliminary Matrix representation in Pyomo kernel. However, a key element of this approach would be to define a Vector representation for parameters and variables. Or to define a convention for using Pyomo's existing Variable/Parameter objects in that manner. But for the hackathon our goal should be to demonstrate the potential performance win.
NOTE: I think it would make sense to include the use of numpy matrix/vector data and scipy.sparse matrix/vector data in our performance hacking. Many users will naturally gravitate to the use of these data formats, and thus I think it would be a mistake to limit the scope of this activity to custom matrix/vector objects.
Related Issues: #685 #468 #31 #87
The text was updated successfully, but these errors were encountered: