When using the Hilbert, for e.g. HOA analysis, the current implementation would allocate numerous identical local buffers (HilbertH/W) and calculate identical coefficients (HilbertH).
Consider optimizations that could minimize the duplication of memory resources and calculations.
- In the case of allocating buffers, this would likely be one layer up from the classes themselves, such as passing in external buffers. This would likely require modification of the current classes.
- In the case of calculating coefficients, one approach is to use a class var to store a dictionary of coefficients for each kernel size, either on demand (built on first request, stored thereafter), or on *initClass for common kernel sizes, and on demand in the case of kernel sizes not pre-calculated.
When using the Hilbert, for e.g. HOA analysis, the current implementation would allocate numerous identical local buffers (HilbertH/W) and calculate identical coefficients (HilbertH).
Consider optimizations that could minimize the duplication of memory resources and calculations.