Alright, we stand to gain MASSIVE, 2x, performance benefit here for all RecurrenceTypes except RecurrenceThreshold. Take a look at our source code that computes the recurrence threshold for a given type: https://github.com/JuliaDynamics/RecurrenceAnalysis.jl/blob/main/src/matrices/recurrence_specification.jl#L78-L153
For all types (besides RecurrenceThreshold) we are computing all the distances across all pairs of points, to estimate a threshold. Then, we give this threshold to the low-level recurrence_matrix function which computes all distances all over again. We can be much smarter than that and just store somewhere the distance matrtix and pass it around until we reach the recurrence_matrix function, which then does a trivial boolean conversion rmat = dmat .< threhold; return SparseMatrix(rmat).
This is such a simple code base improvement with such a massive impact.
Alright, we stand to gain MASSIVE, 2x, performance benefit here for all
RecurrenceTypes exceptRecurrenceThreshold. Take a look at our source code that computes the recurrence threshold for a given type: https://github.com/JuliaDynamics/RecurrenceAnalysis.jl/blob/main/src/matrices/recurrence_specification.jl#L78-L153For all types (besides
RecurrenceThreshold) we are computing all the distances across all pairs of points, to estimate a threshold. Then, we give this threshold to the low-levelrecurrence_matrixfunction which computes all distances all over again. We can be much smarter than that and just store somewhere the distance matrtix and pass it around until we reach therecurrence_matrixfunction, which then does a trivial boolean conversionrmat = dmat .< threhold; return SparseMatrix(rmat).This is such a simple code base improvement with such a massive impact.