Damn Cool Algorithms, Part 1: BK-Trees

http://blog.notdot.net/2007/4/Damn-Cool-Algorithms-Part-1-BK-Trees

by llambda • 15 years ago

177 13 15 years ago

blog.notdot.net

grandpa• 15 years ago

Some implementations:

Lisp: http://cliki.net/bk-tree

Python: http://hupp.org/adam/weblog/?p=103

Ruby: https://github.com/threedaymonk/bktree

Haskell: http://www.robdickerson.net/?p=346

Java: http://code.google.com/p/java-bk-tree/

thesystemis• 15 years ago

here's one in c++ : https://github.com/sholiday/cppBKTree

zszugyi2• 15 years ago

IMHO the explanation on the Lisp implementation page is much clearer then the article.

dman• 15 years ago

If you hang out in the lisp community you will find this to be a recurring theme - the quality of the community and the textbooks makes lisp a great learning language even if you never write a line of lisp code in production. fwiw as a current outsider the haskell community gives me the same feeling that the discourse in the community is of pretty high quality.

omaranto• 15 years ago

What an odd thing to say! The Lisp implementation page has no explanation that I could find of either how to build the tree or how to search it. Or did you mean you found the source code to be a clearer explanation?

Edit: The lisp page does have a nice picture of a tree. Did you mean that looking at that picture you could figure out both the algorithm for building the tree and for querying it and the proof that these are correct faster than by reading Nick's blog post?

rhnet• 15 years ago

I couldn't get the link to the Python version to work, but I wrote my own a while back.

https://github.com/sholiday/pyBKTree

ahupp• 15 years ago

Here's where the python one above (as well as a different haskell version) lives now: https://github.com/ahupp/bktree

leeoniya• 15 years ago

i especially like the part where it says. "The only copy of this online seems to be in the ACM archive, which is subscription only."

kjhughes• 15 years ago

To clarify, Nick Johnson is referring to the original paper on the topic ("Some approaches to best match file searching" by Burkhard and Keller in 1973). It's a common frustration that ACM archives are not freely available to the public.

In the comments, "Sarah" notes that the paper can be found here:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.160...

BTW, Nick's Blog often has gems and is well worth a regular read.

wgd• 15 years ago

This seems similar to the VP-Tree, which there was a post about a few days ago, except that VP-Trees use binary partitioning because they're measuring similarity in a continuous space, while these ones use discrete buckets because similarity is measured discretely.

gujk• 15 years ago

The VP article linked to this one as background material. That is where OP found this article to repost for karma.

arnoldoMuller• 15 years ago

The SISAP metric spaces library comes with an implementation: http://sisap.org/Metric_Space_Library.html You can also compare BK-trees with other methods and you may encounter that their performance is not so great, specially for high dimensional spaces.

Damn Cool Algorithms, Part 1: BK-Trees

Related Stories