Solution to slow subgraph problem :)

Fast-k-cores

require 'graph_client'
require 'pacer'

client = GraphClient.new
k = 16

graph = client.get "louise.kitchen@enron.com"
#k_nodes = graph.v.filter{|v| v.out_e.count > k}.result
#subgraph = k_nodes.out_e.in_v.only(k_nodes).subgraph

puts graph
d = []
graph.v.each do |v|
	if v.out_e.count < k
		d << v
	end
end

d.each do |v|
	graph.remove_vertex v
end

puts graph

As of Pacer 0.8.1 (to be released shortly) you could replace lines 12 to 21 with the following which would be significantly faster.

graph.v.select { |v| v.out_edges.count < k }.delete!

v.out_edges is much faster than v.out_e because out_edges is the native method for an individual vertex which returns an Iterable that then simply gets counted. On the other hand, v.out_e wraps the vertex in a 4 step route and turns that to a pipeline (IdentityPipe -> OutEdgesPipe -> CountSideEffectPipe -> SideEffectCapPipe) that then must be fully iterated. That construction of the route and pipeline isn't terribly heavyweight but inside a loop it is best to avoid it if possible.

Using .delete! on a route is much faster than looping through each of its vertices and calling remove_vertex on each one because it wraps the job using bulk_job so that the entire process can be done in a minimal number of transactions rather than a separate transaction for every vertex plus (potentially) each connected edge for each deleted vertex.

Finally, doing the entire job within the context of a route is much more memory efficient because each vertex is processed as it is emitted from the iterator, so no more than 1 vertex actually needs to be in memory at a time. Much better than collecting everything into an array in advance of processing!

Cheers,
Darrick