search

SCIInstitute / SCIRun

SCIRun is a Problem Solving Environment, for modeling, simulation and visualization of scientific problems. This is version 5, the upgraded version of SCIRun v4.
http://scirun.org
Other
128 stars 71 forks source link

Radial Basis Function solution fail and performance issues #2102

Open wgood opened 4 years ago

wgood commented 4 years ago

Describe the bug I have two point clouds with data, one the data has been altered slightly (has been blurred in matlab). The blurred data fails (the solution is garbage but the module completes), but when you down sample the data it does not. So there doesn't appear to be something wrong with the way the data has been altered. The data is somewhat unique in that it is bound to the bounding surfaces of the destination and that may be why we have not seen this issue previously. I have included the working non-blurred data (Grid_7) the non-working blurred data (Grid_9), and the working downsampled blurred data (Grid_10).

Furthermore, there are performance issues mapping from ~1800 nodes in more extreme cases. This can complete in 10-20 minutes if allowed to run. Although this fails more often than the test case above (and attached) and there does not appear to be a pattern to these failures.

In the below image the solution provided by grid 7 (seemingly correct) and grid 8 (incorrect)

image (16)

Here is the SCIRun network and the necessary data: RBF_testng.zip

Thanks, Wilson

dcwhite commented 4 years ago

Chatted with @jessdtate about this one, and he said it's common with RBF algorithms to encounter this sort of "chaotic," oversensitive behavior. And I'm afraid there's not much we can do to speed up the algorithm, unless it's rewritten to avoid SVD of a large dense matrix. If you find more examples that fail, please attach them.

wgood commented 4 years ago

It appears only cases in which blurring occurred causes the solution to fail. Strangely one of the blurred cases does not run into this issue (grid 10). I have more examples of cases that fail when blurred (~30) but they are blurred in the same way, just with different field data. So let me know if that would be helpful to have. Also, we have not run into this sort of issue with the RBF function before running this on thousands of separate test cases (the solution is so off it is easy to detect when it happens). Although this is the first I am aware of where the data is being blurred before mapping. But there is nothing obvious about the data on my end that would indicate what is causing this issue.

wgood commented 4 years ago

Let me investigate more on my end. I changed a lot of things trying to sort out this issue on my end and that has had some effect on what inputs fail although I have not found one that eliminates all fail cases. Will update soon. Thanks.

github-actions[bot] commented 3 years ago

This issue is stale because it has been open 240 days with no activity. Remove the stale label or comment, or this will be closed in 60 days.

dcwhite commented 3 years ago

@jessdtate Keep this one open? Issue creator is now in another state...

github-actions[bot] commented 2 years ago

This issue is stale because it has been open 240 days with no activity. Remove the stale label or comment, or this will be closed in 60 days.