info.debatty.java.graphs

Class Graph<T>

java.lang.Object

info.debatty.java.graphs.Graph<T>

Type Parameters:

T - The type of nodes value

All Implemented Interfaces:

Serializable

public class Graph<T>
extends Object
implements Serializable

k-nn graph, represented as a mapping node => neighborlist.

Author:

Thibault Debatty

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
static double	DEFAULT_EXPANSION
static int	DEFAULT_K
static int	DEFAULT_LONG_JUMPGS
static double	DEFAULT_SEARCH_SPEEDUP
static int	DEFAULT_UPDATE_DEPTH

Constructor Summary

Constructors

Constructor and Description
Graph() Initialize an empty graph with k = 10.
Graph(int k) Initialize an empty graph, and set k (number of edges per node).

Method Summary

Methods

Modifier and Type	Method and Description
int	add(Node<T> new_node) Add a node to the online graph using exhaustive search approach.
ArrayList<Graph<T>>	connectedComponents() Split the graph in connected components (usually you will first prune the graph to remove "weak" edges).
boolean	containsKey(Node node)
Iterable<Map.Entry<Node<T>,NeighborList>>	entrySet()
int	fastAdd(Node<T> node) Add a node to the online graph, using approximate online graph building algorithm presented in "Fast Online k-nn Graph Building" by Debatty et al.
int	fastAdd(Node<T> node, double speedup) Add a node to the online graph, using approximate online graph building algorithm presented in "Fast Online k-nn Graph Building" by Debatty et al.
int	fastAdd(Node<T> new_node, double speedup, int long_jumps, double expansion) Add a node to the online graph, using approximate online graph building algorithm presented in "Fast Online k-nn Graph Building" by Debatty et al.
int	fastRemove(Node<T> node_to_remove) Remove a node from the graph (and update the graph) using fast approximate algorithm.
NeighborList	fastSearch(T query, int k) Approximate fast graph based search, as published in "Fast Online k-nn Graph Building" by Debatty et al.
NeighborList	fastSearch(T query, int k, double speedup) Approximate fast graph based search, as published in "Fast Online k-nn Graph Building" by Debatty et al.
NeighborList	fastSearch(T query, int k, double speedup, int long_jumps, double expansion) Approximate fast graph based search, as published in "Fast Online k-nn Graph Building" by Debatty et al.
LinkedList<Node<T>>	findNeighbors(LinkedList<Node<T>> starting_points, int depth) Recursively search neighbors of neighbors, up to a given depth.
Node<T>	first() Get the first node in the graph.
NeighborList	get(Node node) Get the neighborlist of this node.
HashMap<Node<T>,NeighborList>	getHashMap() Get the underlying hash map that stores the nodes and associated neighborlists.
int	getK() Get k (the number of edges per node).
Iterable<Node<T>>	getNodes()
SimilarityInterface<T>	getSimilarity() Get the similarity measure.
void	prune(double threshold) Remove from the graph all edges with a similarity lower than threshold.
NeighborList	put(Node<T> node, NeighborList neighborlist)
NeighborList	searchExhaustive(T query, int k) Multi-thread exhaustive search.
void	setDepth(int update_depth) Modify the depth for updating existing edges (default is 2).
void	setK(int k) Set k (the number of edges per node).
void	setSimilarity(SimilarityInterface<T> similarity) Set the similarity measure used to build or search the graph.
void	setWindowSize(int window_size) Set the size of the window (number of nodes to keep in the graph).
int	size()
ArrayList<Graph<T>>	stronglyConnectedComponents() Computes the strongly connected sub-graphs (where every node is reachable from every other node) using Tarjan's algorithm, which has computation cost O(n).
void	writeGEXF(String filename) Writes the graph as a GEXF file (to be used in Gephi, for example).

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

DEFAULT_EXPANSION

public static final double DEFAULT_EXPANSION

See Also:

Constant Field Values

DEFAULT_K

public static final int DEFAULT_K

See Also:

Constant Field Values

DEFAULT_UPDATE_DEPTH

public static final int DEFAULT_UPDATE_DEPTH

See Also:

Constant Field Values

DEFAULT_SEARCH_SPEEDUP

public static final double DEFAULT_SEARCH_SPEEDUP

See Also:

Constant Field Values

DEFAULT_LONG_JUMPGS

public static final int DEFAULT_LONG_JUMPGS

See Also:

Constant Field Values

Constructor Detail

Graph

public Graph(int k)

Initialize an empty graph, and set k (number of edges per node). Default k is 10.

Parameters:

k -

Graph

public Graph()

Initialize an empty graph with k = 10.

Method Detail

getSimilarity

public final SimilarityInterface<T> getSimilarity()

Get the similarity measure.

Returns:

setSimilarity

public final void setSimilarity(SimilarityInterface<T> similarity)

Set the similarity measure used to build or search the graph.

Parameters:

similarity -

getK

public final int getK()

Get k (the number of edges per node).

Returns:

setK

public final void setK(int k)

Set k (the number of edges per node). The existing graph will not be modified.

Parameters:

k -

get

public final NeighborList get(Node node)

Get the neighborlist of this node.

Parameters:

node -

Returns:

the neighborlist of this node

first

public final Node<T> first()
                    throws NoSuchElementException

Get the first node in the graph.

Returns:

The first node in the graph

Throws:

NoSuchElementException - if the graph is empty...

prune

public final void prune(double threshold)

Remove from the graph all edges with a similarity lower than threshold.

Parameters:

threshold -

connectedComponents

public final ArrayList<Graph<T>> connectedComponents()

Split the graph in connected components (usually you will first prune the graph to remove "weak" edges).

Returns:

stronglyConnectedComponents

public final ArrayList<Graph<T>> stronglyConnectedComponents()

Computes the strongly connected sub-graphs (where every node is reachable from every other node) using Tarjan's algorithm, which has computation cost O(n).

Returns:

put

public final NeighborList put(Node<T> node,
               NeighborList neighborlist)

Parameters:

node -

neighborlist -

Returns:

containsKey

public final boolean containsKey(Node node)

Parameters:

node -

Returns:

size

public final int size()

Returns:

entrySet

public final Iterable<Map.Entry<Node<T>,NeighborList>> entrySet()

Returns:

getNodes

public final Iterable<Node<T>> getNodes()

Returns:

findNeighbors

public final LinkedList<Node<T>> findNeighbors(LinkedList<Node<T>> starting_points,
                                int depth)

Recursively search neighbors of neighbors, up to a given depth.

Parameters:

starting_points -

depth -

Returns:

getHashMap

public final HashMap<Node<T>,NeighborList> getHashMap()

Get the underlying hash map that stores the nodes and associated neighborlists.

Returns:

searchExhaustive

public final NeighborList searchExhaustive(T query,
                            int k)
                                    throws InterruptedException,
                                           ExecutionException

Multi-thread exhaustive search.

Parameters:

query -

k -

Returns:

Throws:

InterruptedException

ExecutionException

fastSearch

public final NeighborList fastSearch(T query,
                      int k)

Approximate fast graph based search, as published in "Fast Online k-nn Graph Building" by Debatty et al. Default speedup is 4.

Parameters:

query -

k - search K neighbors

Returns:

See Also:

Fast Online k-nn Graph Building

fastSearch

public final NeighborList fastSearch(T query,
                      int k,
                      double speedup)

Approximate fast graph based search, as published in "Fast Online k-nn Graph Building" by Debatty et al.

Parameters:

query -

k - search k neighbors

speedup - speedup for searching (> 1, default 4)

Returns:

See Also:

Fast Online k-nn Graph Building

fastSearch

public final NeighborList fastSearch(T query,
                      int k,
                      double speedup,
                      int long_jumps,
                      double expansion)

Approximate fast graph based search, as published in "Fast Online k-nn Graph Building" by Debatty et al.

Parameters:

query - query point

k - number of neighbors to find (the K from K-nn search)

speedup - (default: 4.0)

long_jumps - (default: 2)

expansion - (default: 1.2)

Returns:

See Also:

Fast Online k-nn Graph Building

writeGEXF

public final void writeGEXF(String filename)
                     throws FileNotFoundException,
                            IOException

Writes the graph as a GEXF file (to be used in Gephi, for example).

Parameters:

filename -

Throws:

FileNotFoundException - if filename is invalid

IOException - if cannot write to file

setDepth

public final void setDepth(int update_depth)

Modify the depth for updating existing edges (default is 2).

Parameters:

update_depth -

setWindowSize

public final void setWindowSize(int window_size)

Set the size of the window (number of nodes to keep in the graph). Default = 0 = unlimited size

Parameters:

window_size -

add

public final int add(Node<T> new_node)

Add a node to the online graph using exhaustive search approach. Adding a node requires to compute the similarity between the new node and every other node in the graph...

Parameters:

new_node -

Returns:

fastAdd

public final int fastAdd(Node<T> node)

Add a node to the online graph, using approximate online graph building algorithm presented in "Fast Online k-nn Graph Building" by Debatty et al. Default speedup is 4 compared to exhaustive search.

Parameters:

node -

Returns:

fastAdd

public final int fastAdd(Node<T> node,
          double speedup)

Add a node to the online graph, using approximate online graph building algorithm presented in "Fast Online k-nn Graph Building" by Debatty et al. Uses default number of long jumps (2) and default expansion (1.2).

Parameters:

node -

speedup -

Returns:

fastAdd

public final int fastAdd(Node<T> new_node,
          double speedup,
          int long_jumps,
          double expansion)

Add a node to the online graph, using approximate online graph building algorithm presented in "Fast Online k-nn Graph Building" by Debatty et al.

Parameters:

new_node -

speedup - compared to exhaustive search

long_jumps -

expansion -

Returns:

fastRemove

public final int fastRemove(Node<T> node_to_remove)

Remove a node from the graph (and update the graph) using fast approximate algorithm.

Parameters:

node_to_remove -

Returns:

the number of similarities that were computed.