Hi offonoffon... won't the creation of the vertices and edges one at a time end up dramatically affecting the performance of graph creation (see my original question at the top of the thread)? I've found that creating the edges one at a time is MUCH slower than creating them all at once and that creating the vertices one at a time is a little slower. What I have is not pretty though. On 28 April 2015 at 01:11, ... <offonoffoffonoff@gmail.com> wrote:
I solved this once by making a NoDupesGraph where you could add edges with just the names of vertices .
class NoDupesGraph(Graph): '''Add nodes without worrying if it is a duplicate. Add edges without worrying if nodes exist '''
def __init__(self,*args,**kwargs): Graph.__init__(self,*args,**kwargs) self._nodes = {}
def add_nodupe_vertex(self,label,*args,**kwargs): '''Return a node with label. Create node if label is new''' try: n = self._nodes[label] except KeyError: n = self.add_vertex() self._nodes[label]=n return n
def add_nodupe_edge(self, n1_label, n2_label,directed=False): """ Get or create edges using get_or_create_node """ #there may be two if graph is directed but edge isn't edges = []
n1 = self.add_nodupe_vertex(n1_label) n2 = self.add_nodupe_vertex(n2_label) edges.append(self.add_edge(n1,n2)) if self.is_directed() and not directed: edges.append(self.add_edge(n2,n1)) return edges
def flush_empty_nodes(self): '''not implemented''' pass
def condense_edges(self): '''if a node connects to only two edges, combine those edges and delete the node.
not implemented ''' pass
This could be easily modified to suit your need.
On Mon, Apr 27, 2015 at 3:23 PM, Krister <thekswenson@gmail.com> wrote:
Thanks for the quick response Thiago!
In this code all the edges and vertices are created by graph-tool and the result is something much faster... is this the best I can do?
It's somewhat annoying to have to keep track of the vertices that will be created like this:
def graph_tool_create_all_at_once(): """ Create a graph_tool graph given a list of pairs. """ G = Graph(directed=False) objectTOi = {} vertexpairs = [] counter = 0 for o1,o2 in get_pairs_of_ints(): if(o1 in objectTOi): u = objectTOi[o1] else: u = counter counter += 1 objectTOi[o1] = u if(o2 in objectTOi): v = objectTOi[o2] else: v = counter counter += 1 objectTOi[o2] = v
vertexpairs.append((u,v))
G.add_edge_list(vertexpairs)
On 27 April 2015 at 16:39, Tiago de Paula Peixoto <tiago@skewed.de> wrote:
On 27.04.2015 14:29, thekswenson wrote:
I've been using networkx to simply create a graph and check the connected components. The bottleneck of the operation is the creation of the edges.
I've heard that graph-tool is very efficient so I've replaces the code with a graph-tool graph. To my surprise, the creation of a graph-tool graph is MUCH slower than that of a networkx graph.
Am I doing something wrong?
How does the performance change if you create the necessary edges beforehand?
In graph-tool things are faster than in networkx when they are delegated to C++, otherwise this should be comparable in speed. In the case of adding many edges, this is done by using the Graph.add_edge_list() function, which runs in C++ internally. In your example, this should provide a massive speed-up.
Best, Tiago
-- Tiago de Paula Peixoto <tiago@skewed.de>
_______________________________________________ graph-tool mailing list graph-tool@skewed.de http://lists.skewed.de/mailman/listinfo/graph-tool
_______________________________________________ graph-tool mailing list graph-tool@skewed.de http://lists.skewed.de/mailman/listinfo/graph-tool
_______________________________________________ graph-tool mailing list graph-tool@skewed.de http://lists.skewed.de/mailman/listinfo/graph-tool