[mlpack-git] [mlpack/mlpack] Implementation of Multiprobe LSH (#691)

[Yannis Mentekidis]

So, there is an error in my logic behind MultiprobeDeterministicTest. I think it's an error in the test and not in the code. Maybe it would be better to remove that test completely. Here's where I went wrong.

When creating the four clusters of points, what I had in mind was this:
![PointCluster.png](http://i.imgur.com/JcozL2s.png)

So the groups of blue points would be assigned the same code. This doesn't happen because of the random offsets added before flooring, so the actual image is more like:
![PointCluster_offset.png](http://i.imgur.com/VJggrVB.png)

meaning, each cluster is not confined to only one code as I initially thought. This makes this test fail, because for example the red (query) point will have neighbors even without multiprobe but will obviously have more neighbors when multiprobe is used. This is something we actually check in the randomized check, so I removed this test until I could find a better one.

Note that the single-probe deterministic test still works - the clusters don't mix

---
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/mlpack/mlpack/pull/691#issuecomment-228315339
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.cc.gatech.edu/pipermail/mlpack-git/attachments/20160624/45779252/attachment-0001.html>