jizhihang / spatialtree Goto Github PK

spatialtree: Python module for spatial trees
Author:     Brian McFee <[email protected]>
CREATED:    2011-11-13 16:12:29 

This code is distributed under the GNU GPL license.  See LICENSE for details, 
or http://www.gnu.org/licenses/gpl-3.0.txt .

If you use this code for academic research, please cite the following 
publication:

[1] McFee, B. and Lanckriet, G.R.G.  Large-scale music similarity search 
    with spatial trees.  12th International Society for Music Information 
    Retrieval (ISMIR) conference, 2011.

INTRODUCTION
------------

This module provides a unified interface to constructing various flavors of 
spatial tree data structures for accelerating approximate nearest-neighbor
retrieval in high-dimensional data.

The supported methods for generating spatial trees include:
    * KD-trees (maximum-variance) [2]
    * PCA-trees [3]
    * 2-means trees [3]
    * Random projection trees [4]

The methods listed above provide different rules for generating a recursive 
partitioning of high-dimensional vector data.  The spatialtree package also
provides support for spill trees, which use redundant mappings to improve
the accuracy of nearest-neighbor retrieval [5].  The spill tree functionality 
may be combined with any of the above rules.

Spatialtree supports indexing of raw vector/matrix data (in the form of numpy 
arrays), or structured key-value stores.  Spatialtree is semi-dynamic, in 
that the tree may be pruned to a fixed height, and data may be removed (and 
added, if using key-value stores), but the tree does not re-balance.

For static data sets, we provide an efficient and light-weight inverted map 
data structure for answering (approximate) nearest neighbor queries of items
within the set.

Several example programs are provided, demonstrating the various use-cases.
Class and method documentation is provided in doc-strings (pydoc spatialtree).


INSTALLATION
------------

From the command-line (as root/sudo):

# python setup.py install


REQUIREMENTS
------------

This module depends on numpy and scipy.

REFERENCES
----------

[2] J.L. Bentley. Multidimensional binary search trees used for
    associative searching. Commun. ACM, 18:509–517, Sep. 1975.

[3] Nakul Verma, Samory Kpotufe, and Sanjoy Dasgupta. Which
    spatial partition trees are adaptive to intrinsic dimension? In
    Uncertainty in Artificial Intelligence, pages 565–574, 2009.

[4] Sanjoy Dasgupta and Yoav Freund. Random projection trees
    and low dimensional manifolds. In ACM Symposium on Theory
    of Computing, pages 537–546, 2008.

[5] Ting Liu, Andrew W. Moore, Alexander Gray, and Ke Yang.
    An investigation of practical approximate nearest neighbor 
    algorithms. In NIPS, pages 825–832. 2005.