I would like to know if somebody has already coded functions for creating
some small-world / scale free / ER networks with the following parameters
(similar to networkx):
small-world -> n : The number of nodes , k : Each node is connected to k
nearest neighbors in ring topology , p: The probability of rewiring each
edge
scale-free -> n : Number of nodes, m : Number of edges to attach from a new
node to existing nodes
ER -> n : The number of nodes , p : Probability for edge creation.
small-world -> n : The number of nodes , k : Each node is connected to
k nearest neighbors in ring topology , p: The probability of rewiring
each edge
There no function for this yet. It is in my TODO list. But you can
create this easily by creating a 1D lattice (a ring) with the lattice()
function, and than randomly rewire a portion of the edges.
scale-free -> n : Number of nodes, m : Number of edges to attach from
a new
node to existing nodes
This is implemented in the price_network() function.
ER -> n : The number of nodes , p : Probability for edge creation.
This is also in my TODO list, although it is very easy to
implement. Note that, for most purposes, this is equivalent to
generating a random graph with random_graph() and using a poisson degree
distribution with the appropriate average.
I have followed your advice for the scale-free and er network and there are
no pb for these.
The watts-strogatz network version I have implemented works, but is very
very slow.
See code:
def watts_strogatz_network(n,k,p):
g=lattice([n])
# make a ring
g.add_edge(g.vertex(0), g.vertex(n-1))
# add edges to k neighbours of each vertex in the ring
# k-1 neighbours if k is odd
if ((k%2!=0) and k>1):
k-=1
if k>2:
for v in range(n):
l_n=get_n_levels_neighbours(n,v,k)
for v_n in l_n:
if not g.edge(v_n,v):
g.add_edge(g.vertex(v), g.vertex(v_n))
# replace each edge u-v by an edge u-w with probability p
for u in range(n):
for v in g.vertex(u).all_neighbours():
if (random()<=p):
l1=range(n)
l2=get_n_levels_neighbours(n,u,k)
l=[i for i in l1 if i not in l2]
l.remove(u) #print l
w=rd.choice(l)
while w==u or g.edge(u,w):
w=rd.choice(l)
g.remove_edge(g.edge(u, v))
g.add_edge(g.vertex(u), g.vertex(w))
return g
def get_n_levels_neighbours(maxn,n,k):
l=range(n-(k/2),n)
l+=range(n+1,n+(k/2)+1)
for i in range(len(l)):
if (l[i]<0):
l[i]=((l[i])%(maxn-1))+1
if(l[i]>(maxn-1)):
l[i]=l[i]%maxn
return l
I have followed your advice for the scale-free and er network and
there are no pb for these. The watts-strogatz network version I have
implemented works, but is very very slow.
[...]
Yes, indeed. Doing it in C++ would be much faster.
But there are ways to improve your code:
- Do not call the range() function, since it is O(N). In loops, you
should call xrange() instead.
- To sample a random vertex, you can do something like:
g.vertex(numpy.random.randint(g.num_vertices()))
This is much faster than using choice().
I have noticed that the boost library has a small-world graph
generator see: Boost Graph Library: Small World Generator - 1.40.0
Would there be any way to call that inside graph-tool as a faster
alternative?
Yes, I will implement this as soon as I have some time.