18807600203 / lsh-hadoop Goto Github PK

This project implements a new Locality-Specific Hashing technique for nearest-neighbor problems.
http://en.wikipedia.org/wiki/Locality_sensitive_hashing
http://en.wikipedia.org/wiki/Nearest_neighbor_search

This paper explains the theory of this implementation:
Abstract: 
http://www.sciweavers.org/publications/locality-sensitive-hash-real-vectors

Full paper: 
http://siam.org/proceedings/soda/2010/SODA10_094_neylont.pdf

The code for the LSH algorithm is all courtesy of examples and explanation from Tyler.

The project has three parts: 
lsh.ortho: 	the core implementation 
lsh.solr:	Solr "reprocessor": download all docs, find neighbors, 
		upload 
		Map/reduce assistance for Solr doc m/r jobs
		Contains an M/R algorithm that adds corners to all 
		documents.
		Both are tested with NY State cinema location data.
lsh.hadoop:	map/reduce wrapper. 
side utility:	Hadoop file reader which parses input csv lines and rearranges them.


The core insight is to use equilateral triangles instead of squares to round off locations. That's it.