LSHKIT: A C++ Locality Sensitive Hashing Library

This library is part of the CASS project.

Wei Dong wdong [@]



lshkit 0.2.1 - 2009-09-17

* A Posterirori Multi-Probe LSH (include/lshkit/apost.h tools/apost-run.cpp)
* Spectral hashing (include/lshkit/spectral-hash.h)

lshkit pre-0.2 - 2009-03-02

* Introduced experimental scanner interface.  Scanner is a unary functional object which is passed into
LSH index to scan the candidate points and update the K-NN results.  There are several benefits of
this change:
1. The LSH index query method are now thread-safe (if scanners are thread-safe).  This makes parallel
query execution possible.
2. It makes implementation simpler.  LSH index classes no longer need to maintain accessor and metric
3. Without changing the LSH index API, one can now easily combine LSH indexing and sketch filtering
(using skeches to quickly filter the LSH candidate, and then use raw data to rank reduced candidate set).
This will further improve the query speed when LSH return a large number of candidates.  An example
will be provided shortly.

lshkit 0.1.9 - 2009-03-02

* Improved the MPLSH modeling algorithm to make it work with feature vectors with large absolute values.
* Added a preliminary LSH Forest implementation with an example program tools/forest-run.cpp.
* Added tools/embedder.cpp, a sample program of random histogram embedding.
* Added R-NN (range query) support to Topk.  
* Added R-NN support in tools/scan.cpp, tools/mplsh-run.cpp, tools/lsh-run.cpp, tools/sketch-run.cpp.  Set K = 0 for R-NN.
* Added tools/txt2bin.cpp to convert text data file to binary.

The current index implementation is not thread-safe.  I'm going to change the index API in next version (0.2).  This is a major change.  This is the last release with the old index API.

lshkit 0.1.2 - 2009-02-15

* Minor fix of the included boost distribution, no LSHKIT file touched.

lshkit 0.1.1 - 2009-02-14
* Added the required Boost source files to the package, so a standalone boost installation is not needed.
* Reduced dependency on Boost.  Now only boost_program_option needs to be linked for the tools. The main LSHKIT library only depends on Boost header files.

lshkit 0.1 - 2009-02-11

* This is the initial versioned release of LSHKIT.
* The project is migrated to sourceforge.