Opened 19 years ago
Closed 19 years ago
#4 closed enhancement (fixed)
improve efficency of least_squares algorithm for speed increase
Reported by: | ole | Owned by: | duncan |
---|---|---|---|
Priority: | normal | Milestone: | least_squares module finished |
Component: | Efficiency and optimisation | Version: | |
Severity: | normal | Keywords: | |
Cc: | ole |
Description (last modified by duncan)
Arrange quad tree algorithm so that all datapoints contribute to interpolation matrix A, without having to set the expand search flag to true.
Currently this is fixed by expand search but that can be very slow. Need e.g. algorithm for searching neighbouring treenodes.
Change History (9)
comment:1 Changed 19 years ago by ole
- Milestone changed from AnuGA ready for release to least_squares module finished
comment:2 Changed 19 years ago by ole
- Description modified (diff)
comment:3 Changed 19 years ago by ole
- Description modified (diff)
comment:4 Changed 19 years ago by duncan
- Description modified (diff)
- Summary changed from least_squares fit does not always use all datapoints to improve efficency of least_squares algorithm for speed increase
comment:5 Changed 19 years ago by duncan
- Severity changed from normal to enhancement
comment:6 Changed 19 years ago by ole
- Component changed from Compilation and installation to Efficiency and optimisation
comment:7 Changed 19 years ago by duncan
- Status changed from new to assigned
comment:8 Changed 19 years ago by duncan
Previous comment should've said interpolate, not fit.
comment:9 Changed 19 years ago by duncan
- Resolution set to fixed
- Status changed from assigned to closed
The same search algorithm is used. Cropping is now inforced, which means expand serach is not slow.
Note: See
TracTickets for help on using
tickets.
The slowness occurs when expand_search is True and cropping is False. This can happen when you want to fit a mesh to grid points. There will be no values to the grid points outside of the mesh.
A solution would be to always crop and return the cropped points, so they aren't lost.