[mlpack-git] (blog) master: Move old post from mail list to the blog. (23c1382)

[gitdub at mlpack.org]

Repository : https://github.com/mlpack/blog
On branch  : master
Link       : https://github.com/mlpack/blog/compare/6ff3ad351ee2af8542bd0eb4474f815bfbf2a036...f91dcc036add6ffbfa9000caa1b6d8a90e25f332

>---------------------------------------------------------------

commit 23c13827785bdd7008732b1ec4b8077e499a59e5
Author: MarcosPividori <marcos.pividori at gmail.com>
Date:   Mon Jun 6 08:28:08 2016 -0300

    Move old post from mail list to the blog.


>---------------------------------------------------------------

23c13827785bdd7008732b1ec4b8077e499a59e5
 content/blog/MarcosWeekOne.md | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/content/blog/MarcosWeekOne.md b/content/blog/MarcosWeekOne.md
new file mode 100644
index 0000000..1de79a0
--- /dev/null
+++ b/content/blog/MarcosWeekOne.md
@@ -0,0 +1,11 @@
+Title: Approximate Nearest Neighbor Search
+Date: 2016-05-29 21:00:00
+Tags: gsoc, knn, kfn
+Author: Marcos Pividori
+
+I am working this summer on the implementation of approximate nearest neighbor search.
+First, we want to update the existing dual tree algorithm to do approximate nearest neighbor search, modifying the prune rule to include an epsilon, as mentioned at the end of the paper [1](http://www.ratml.org/pub/pdf/2015faster.pdf).
+After reading the paper: "Tree-Independent Dual-Tree Algorithms" [2](http://www.ratml.org/pub/pdf/2013tree.pdf) in detail, I found some issues in the definition of the B2 bound. We were discussing about it in [3](http://github.com/mlpack/mlpack/issues/642) and, after thinking about the different special cases, we found some examples where actual definition could fail. We seem to have a solution. New week, I will work in updating existing code to fix this and compare the performance after this modification.
+When reviewing the implementation of tree types, I found a subtle mistake! Ball trees were using hyper rectangle bounds instead of ball bounds! [4](http://github.com/mlpack/mlpack/pull/646) It was fixed and is waiting to be merged.
+
+Next week, I plan to continue improving the neighbor search code, to fix the issue related to the bounds. Once this is working properly, I could add the modification to do approximate search and do some tests.