igraph a package for network analysis G a bor Cs a rdi - - PowerPoint PPT Presentation

igraph – a package for network analysis

G´ abor Cs´ ardi

Gabor.Csardi@unil.ch Department of Medical Genetics, University of Lausanne, Lausanne, Switzerland

Outline

• 1. Why another graph package?
• 2. igraph architecture, data model and data representation
• 3. Manipulating graphs
• 4. Features and their time complexity

igraph – a package for network analysis 2

Why another graph package?

• graph is slow. RBGL is slow, too.

?

1

> ba2 # graph & RBGL

2

A graphNEL graph with undirected edges

3

Number of Nodes = 100000

4

Number of Edges = 199801

Why another graph package?

• graph is slow. RBGL is slow, too.

?

1

> ba2 # graph & RBGL

2

A graphNEL graph with undirected edges

3

Number of Nodes = 100000

4

Number of Edges = 199801

5

> system.time(RBGL::transitivity(ba2))

6

user system elapsed

7

7.517 0.000 7.567

Why another graph package?

• graph is slow. RBGL is slow, too.

?

1

> ba2 # graph & RBGL

2

A graphNEL graph with undirected edges

3

Number of Nodes = 100000

4

Number of Edges = 199801

5

> system.time(RBGL::transitivity(ba2))

6

user system elapsed

7

7.517 0.000 7.567

8

> summary(ba) # igraph

9

Vertices: 1e+05

10

Edges: 199801

11

Directed: FALSE

12

No graph attributes.

13

No vertex attributes.

14

No edge attributes.

Why another graph package?

• graph is slow. RBGL is slow, too.

?

1

> ba2 # graph & RBGL

2

A graphNEL graph with undirected edges

3

Number of Nodes = 100000

4

Number of Edges = 199801

5

> system.time(RBGL::transitivity(ba2))

6

user system elapsed

7

7.517 0.000 7.567

8

> summary(ba) # igraph

9

Vertices: 1e+05

10

Edges: 199801

11

Directed: FALSE

12

No graph attributes.

13

No vertex attributes.

14

No edge attributes.

15

> system.time(igraph::transitivity(ba))

16

user system elapsed

17

0.328 0.000 0.335

igraph – a package for network analysis 3

Why another graph package?

• sna is slow. network is slow, too.

1

> net2 # SNA & network

2

Network attributes:

3

vertices = 1e+05

4

directed = TRUE

5

hyper = FALSE

6

loops = FALSE

7

multiple = FALSE

8

bipartite = FALSE

9

total edges= 199801

10

missing edges= 0

11

non-missing edges= 199801

12

...

Why another graph package?

• sna is slow. network is slow, too.

1

> net2 # SNA & network

2

Network attributes:

3

vertices = 1e+05

4

directed = TRUE

5

hyper = FALSE

6

loops = FALSE

7

multiple = FALSE

8

bipartite = FALSE

9

total edges= 199801

10

missing edges= 0

11

non-missing edges= 199801

12

...

13

> gtrans(net2)

14

Error in matrix(0, nr = network.size(x), nc = network.size(x)) :

15

too many elements specified

igraph – a package for network analysis 4

Why another graph package?

• graph is slow. RBGL is slow, too.
• sna is slow. network is slow, too.
• A generic solution was needed, i.e. a common C layer, that can be interfaced from

C/C++, R, Python, etc.

Why another graph package?

• graph is slow. RBGL is slow, too.
• sna is slow. network is slow, too.
• A generic solution was needed, i.e. a common C layer, that can be interfaced from

C/C++, R, Python, etc.

graph library core GNU R Python Ruby

igraph – a package for network analysis 5

The igraph architecture

API igraph_t API utility types igraph library R glue layer R functions converter

igraph – a package for network analysis 6

Dependencies

• Standard C/C++ libraries.

Dependencies

• Standard C/C++ libraries.
• stats package, this is part of base.

Dependencies

• Standard C/C++ libraries.
• stats package, this is part of base.
• Optional: libxml2 library, for reading

GraphML files (included in Windows builds).

Dependencies

• Standard C/C++ libraries.
• stats package, this is part of base.
• Optional: libxml2 library, for reading

GraphML files (included in Windows builds).

• Optional: GMP library, graph automorphisms

(not included in Windows builds).

Dependencies

• Standard C/C++ libraries.
• stats package, this is part of base.
• Optional: libxml2 library, for reading

GraphML files (included in Windows builds).

• Optional: GMP library, graph automorphisms

(not included in Windows builds).

• Suggested packages: stats4, rgl, tcltk, RSQLite,

digest, graph, Matrix.

igraph – a package for network analysis 7

The igraph data model, what cannot be represented

“Mixed”graphs, with undirected and directed edges. You can“emulate”them via graph attributes.

The igraph data model, what cannot be represented

“Mixed”graphs, with undirected and directed edges. You can“emulate”them via graph attributes.

• Hypergraphs. Perhaps see the hypergraph package.

The igraph data model, what cannot be represented

“Mixed”graphs, with undirected and directed edges. You can“emulate”them via graph attributes.

• Hypergraphs. Perhaps see the hypergraph package.

No direct support for bipartite (two-mode) graphs. It is possible to handle them via graph attributes.

igraph – a package for network analysis 8

Graph representation, sparse graphs

Flat data structures, indexed edge lists. Easy to handle, good for many kind of questions.

igraph – a package for network analysis 9

Graph representation, sparse graphs

Flat data structures, indexed edge lists. Easy to handle, good for many kind of questions.

igraph – a package for network analysis 10

Graph representation, sparse graphs

Flat data structures, indexed edge lists. Easy to handle, good for many kind of questions.

igraph – a package for network analysis 11

Graph representation, sparse graphs

Flat data structures, indexed edge lists. Easy to handle, good for many kind of questions.

igraph – a package for network analysis 12

Graph representation, sparse graphs

Flat data structures, indexed edge lists. Easy to handle, good for many kind of questions.

X

A B C D E F G H I J

igraph – a package for network analysis 13

Graph representation, sparse graphs

Flat data structures, indexed edge lists. Easy to handle, good for many kind of questions.

X

A B C D E F G H I J

X

A B C D E F G H I J

igraph – a package for network analysis 14