def test_grid_graph():
graph = nx.grid_graph([20,20])
MC_steps = 1000000
true_sample_trials = 1
samples = run_markov_chain(graph, MC_steps, .1)
#for sample in true_samples:
# viz(graph, sample[0], sample[1])
viz(graph, samples[-10][1], samples[-10][2])
#The question: Can we tune temp so that we get both rapid mixing AND polynomial concentration on the zero contradiction assignments??
#We can falsify rapid mixing using the totally uniform samples that we have.
non_cut_sets = [len(s[1]) for s in samples]
sample_num_components = [len (list(
nx.connected_components( nx.edge_subgraph(graph, x[1])) )) for x in samples]
true_samples = test_rejection_sample(graph, true_sample_trials)
true_non_cut_sets = [len(x[0]) for x in true_samples]
true_num_components = [len (list( nx.connected_components( nx.edge_subgraph(graph, x[0])) )) for x in true_samples]
print(np.mean(non_cut_sets), " vs. ", np.mean(true_non_cut_sets))
for sample in true_samples:
viz(graph, sample[0], sample[1])
print(np.mean(sample_num_components), " vs. ", np.mean(true_num_components))
def test_edge_subgraph(self):
assert (self.G.edge_subgraph([
(1, 2), (0, 3)
]).adj == nx.edge_subgraph(self.G, [(1, 2), (0, 3)]).adj)
assert (self.DG.edge_subgraph([
(1, 2), (0, 3)
]).adj == nx.edge_subgraph(self.DG, [(1, 2), (0, 3)]).adj)
def load_graph(a):
global G
if (a==0):
try:
G = ox.load_graphml('network.graphml') #load
print("loaded from graphml file")
except: pass
elif (a==1):
try:
G = ox.graph_from_xml('network.osm', bidirectional=False, simplify=True, retain_all=False)
e = list(G.edges(data=True))
sub_ed = []
for el in e:
if 'highway' in el[2]:
if (el[2]['highway'] in ['primary','motorway','trunk','secondary','tertiary','residential','road','unclassified']):
sub_ed.append(tuple([el[0],el[1],0]))
G = nx.edge_subgraph(G,sub_ed)
print("loaded from osm xml")
except: pass
if (G is None):
try:
G = ox.graph_from_place('Ижевск, Россия', network_type='drive',simplify=True)
print("downloaded from internet")
except: pass
if (G is None):
print('failed to load map. To download network from internet check internet connection. To load from file it\'s name should be \'network.graphml\' or \'network.xml\'')
raise 0
return
def retrieve_topo_from_ONOS(self):
logging.info("Retrieving Topology...")
reply = json_get_req('http://%s:%d/onos/v1/devices' % (ONOS_IP, ONOS_PORT))
if 'devices' not in reply:
return
devices = [dev['id'] for dev in reply['devices'] if dev['available']]
self.G.remove_nodes_from([n for n in self.G if n not in set(devices)])
self.G.add_nodes_from(devices)
reply = json_get_req('http://%s:%d/onos/v1/hosts' % (ONOS_IP, ONOS_PORT))
if 'hosts' not in reply:
return
for host in reply['hosts']:
for location in host['locations']:
self.hosts[host['id']] = location['elementId']
reply = json_get_req('http://%s:%d/onos/v1/links' % (ONOS_IP, ONOS_PORT))
if 'links' not in reply:
return
edges_set = set([(el['src']['device'], el['dst']['device']) \
for el in reply['links'] if el['state'] == 'ACTIVE'])
for edge in itertools.permutations(self.G.nodes(), 2):
is_edge_active = edge in edges_set
self.G.add_edge(*edge, **{ 'active':is_edge_active })
self.update_edge_metrics(is_edge_active, edge)
self.topo = nx.edge_subgraph(self.G, edges_set)
logging.debug("Edge set:", edges_set)
def _set_contagion_subgraph(self):
edge_set = set()
for edge in self.graph.edges:
if self.graph.edges[edge]['used_time'] == 1:
edge_set.add(edge)
self.contagion_subgraph = nx.Graph(
nx.edge_subgraph(self.graph, edge_set))
def plot_networkx(G, ax=None, only_true=False, edge_feature='solution', threshold=0.5):
"""G is networkx graph,
node feature: {'pos': [r, phi, z]}
edge feature: {"solution": []}
"""
if ax is None:
_, ax = plt.subplots()
n_edges = len(G.edges())
edge_colors = [0.]*n_edges
true_edges = []
for iedge,edge in enumerate(G.edges(data=True)):
if np.isscalar(edge[2][edge_feature]):
score = edge[2][edge_feature]
else:
score = edge[2][edge_feature][0]
if score > threshold:
edge_colors[iedge] = 'r'
true_edges.append((edge[0], edge[1]))
else:
edge_colors[iedge] = 'grey'
Gp = nx.edge_subgraph(G, true_edges) if only_true else G
edge_colors = ['k']*len(true_edges) if only_true else edge_colors
pos = get_pos(Gp)
nx.draw(Gp, pos, node_color='#A0CBE2', edge_color=edge_colors,
width=0.5, with_labels=False, node_size=1, ax=ax, arrows=False)
def get_true_subgraph(G):
true_edges = []
for iedge, edge in enumerate(G.edges(data=True)):
if int(edge[2]['solution'][0]) == 1:
true_edges.append((edge[0], edge[1]))
Gp = nx.edge_subgraph(G, true_edges)
return Gp
def edge_type_subgraph(graph, edge_type):
edge_types = []
if isinstance(edge_types, list):
edge_types = edge_type
else:
edge_types = [edge_type]
subg = nx.edge_subgraph(
graph, [(e[0], e[1])
for e in graph.edges().data() if e[2]['type'] in edge_types])
return subg
def extract_year_mgs(mg):
year_mgs = {}
for year in years:
select_edges = []
for (u, v, k, d) in mg.edges(data=True, keys=True):
edge_year = d["start"]
if edge_year <= year:
select_edges.append((u, v, k))
year_mg = nx.edge_subgraph(mg, select_edges)
year_mgs[year] = year_mg
return year_mgs
def setup(self):
# Create a path graph on five nodes.
self.G = G = nx.path_graph(5)
# Add some node, edge, and graph attributes.
for i in range(5):
G.nodes[i]['name'] = 'node{}'.format(i)
G.edges[0, 1]['name'] = 'edge01'
G.edges[3, 4]['name'] = 'edge34'
G.graph['name'] = 'graph'
# Get the subgraph induced by the first and last edges.
self.H = nx.edge_subgraph(G, [(0, 1), (3, 4)])
def setup(self):
# Create a path graph on five nodes.
self.G = G = nx.path_graph(5)
# Add some node, edge, and graph attributes.
for i in range(5):
G.nodes[i]['name'] = 'node{}'.format(i)
G.edges[0, 1]['name'] = 'edge01'
G.edges[3, 4]['name'] = 'edge34'
G.graph['name'] = 'graph'
# Get the subgraph induced by the first and last edges.
self.H = nx.edge_subgraph(G, [(0, 1), (3, 4)])
def setup_class(cls):
# Create a path graph on five nodes.
cls.G = G = nx.path_graph(5)
# Add some node, edge, and graph attributes.
for i in range(5):
G.nodes[i]["name"] = f"node{i}"
G.edges[0, 1]["name"] = "edge01"
G.edges[3, 4]["name"] = "edge34"
G.graph["name"] = "graph"
# Get the subgraph induced by the first and last edges.
cls.H = nx.edge_subgraph(G, [(0, 1), (3, 4)])
def eval_recuit(sol, graph, terms):
#arretes du graphe
edges = list(nx.edges(graph))
#arretes du sous graphe
edgesSub = []
#on selectionne les arretes à 1 dans 'sol'
for i in range(len(edges)):
if (sol[i]):
edgesSub.append(edges[i])
#sous graphe à partir des arretes selectionees
subGraph = nx.edge_subgraph(graph, edgesSub)
#somme des poids de ce sous-graphe
sumW = subGraph.size(weight="weight")
#nombre de termes non-reliés
nbTermsNR = 0
#termes présents dans le sous-graphe
sg_terms = []
for t in terms:
if (subGraph.has_node(t)):
sg_terms.append(t)
else:
#print(t," is not in subGraph ! ")
# si le noeud n'est pas dans le sous-graphe alors il n'est pas relié
nbTermsNR += 1
linked = False
for t1 in sg_terms:
for t2 in sg_terms:
if t1 != t2 and nx.has_path(subGraph, t1, t2):
#print(t1,"is linked with",t2)
linked = True
break
if not linked:
nbTermsNR += 1
#print_graph(subGraph, sg_terms)
#print(nbTermsNR,"node not linked")
Mt = graph.size()
Mc = graph.size() / 2
nbCC = len(list(nx.connected_components(subGraph)))
res = sumW + Mt * nbTermsNR + Mc * (nbCC - 1)
#print("sumW :",sumW,"NR :",nbTermsNR,"nbCC :",nbCC,"eval :",res)
return res
def check_simple_cycle(graph, edges):
edge_list = []
for x in edges:
edge = []
for e in x:
edge.append(e)
edge = tuple(edge)
edge_list.append(edge)
H = nx.edge_subgraph(graph, edge_list)
if len(H) == 0:
return True
if not nx.is_connected(H):
return False
if np.max(list(dict(nx.degree(H)).values())) > 2:
return False
return True
def drawSolGraph(graph, terms, sol):
#arretes du graphe
edges = list(nx.edges(graph))
#arretes du sous graphe
edgesSub = []
#on selectionne les arretes à 1 dans 'sol'
for i in range(len(edges)):
if (sol[i]):
edgesSub.append(edges[i])
#sous graphe à partir des arretes selectionees
subGraph = nx.edge_subgraph(graph, edgesSub)
sg_terms = []
for t in terms:
if (subGraph.has_node(t)):
sg_terms.append(t)
print_graph(subGraph, sg_terms)
def __call__(self):
G = self.G
n = len(G)
# Get list of edges with positive value in the current LP solution
elist = [e for e in G.edges if self.get_values(G[e[0]][e[1]]['var'])>1.0E-5]
# If with animation, show the (fractional) solution
if self.anim:
objv = self.get_objective_value()
xval = [ self.get_values( G[e[0]][e[1]]['var'] ) for e in elist ]
efrac = [ e for e in elist if self.get_values(G[e[0]][e[1]]['var'])< 0.99 ]
plotEdgeList(self.problem, elist, specialEdges=efrac,\
title="Fractional LP solution of objval="+str(objv) )
# Build the solution's support graph, i.e. the sub-graph induced by
# the edges in the list "elist" found above.
supG = nx.edge_subgraph(G, elist)
# If the support graph G is not connected, build the SCE constraint
# based on the graph's smallest component.
S = min( nx.connected_components(supG), key=len )
weight = 1.0
# If the graph is connected, obtain the minimum-weight cut set
# using the algorithm of Stoer and Wagner
if len(S)==n:
# Set the edge weights equal to the variable's solution values
for e in elist:
i, j = e[0], e[1]
supG[i][j]['weight'] = self.get_values( G[i][j]['var'] )
weight, part = nx.stoer_wagner(supG)
S = part[0] if len(part[0]) < len(part[1]) else part[1]
# If the cutset constraint is violated, include the cut in
# form of the equivalent subcycle elimination constraint
if ( len(S) > 2 ) and ( weight < 1.98):
E = list(combinat(S,2))
cycVars = [ G[e[0]][e[1]]['var'] for e in E ]
self.add( cplex.SparsePair(cycVars, [1.0]*len(E) ), sense="L", \
rhs=len(S)-1)
def build_adj_lcc(args):
"""Build the mutual mention largest connected component from
a csv edgelist with weights."""
logging.info("Read '%s' into an adjacency matrix",
args.weighted_edgelist_path)
adj_orig, uid2orig = csv_to_adj(args.weighted_edgelist_path)
logging.info("Convert adjacency matrix to a graph")
g_orig = nx.from_scipy_sparse_matrix(adj_orig,
create_using=nx.DiGraph,
edge_attribute='weight')
logging.info("Extract mutual mention graph from full graph")
g_mutual = nx.edge_subgraph(g_orig, [
e for e in g_orig.edges
if is_edge_mutual(g_orig, e, args.mutual_threshold)
])
g_mutual = nx.Graph(g_mutual)
logging.info("Extract largest connected component from "
"mutual mention graph")
g_lcc = g_mutual.subgraph(max(nx.connected_components(g_mutual), key=len))
logging.info("Relabel largest connected component node ids")
orig2lcc = dict((id, n) for n, id in enumerate(g_lcc.nodes))
g_lcc = nx.relabel_nodes(g_lcc, orig2lcc)
assert set(g_lcc.nodes) == set(range(len(g_lcc.nodes))), \
"Mutual mention lcc does not cover all node ids"
logging.info("Convert largest connected component to adjacency matrix")
adj_lcc = nx.to_scipy_sparse_matrix(g_lcc,
dtype=np.uint32,
weight='weight',
nodelist=sorted(g_lcc.nodes()))
adj_lcc = coo_matrix(adj_lcc)
logging.info("Number of nodes in the largest connected component = %s",
len(g_lcc.nodes))
return adj_lcc, uid2orig, orig2lcc
def get_tracks2(G, th=0.5, feature_name='solution'):
used_nodes = []
sub_graphs = []
for node in G.nodes():
if node in used_nodes:
continue
a_track = longest_track(G,
node,
used_nodes,
th=th,
feature_name=feature_name)
if len(a_track) < 1:
used_nodes.append(node)
continue
sub = nx.edge_subgraph(G, a_track)
sub_graphs.append(sub)
used_nodes += list(sub.nodes())
n_tracks = len(sub_graphs)
print("total tracks:", n_tracks)
return sub_graphs
def generate_unions_of_paths(n=4, l=3, r=3, random=False):
graph = complete_layered_digraph(n, l)
paths = nx.all_simple_paths(graph, graph.graph["s"], graph.graph["t"])
'''
We could do a backtraking tree, where we put edges either into J (the edge set
), or its complement. Checking whether such an assignment can be completed amounts to verifying
that the complement is not a cut of the layered complete digraph, so we can delete and
ask networkx for connectivity. In some cases it may be that certain sets of edges
will necessarily be included, but maybe sufficiently efficient to find them on later levels of the tree.
Of course there are still a lot of connected layered digraphs. ... actually most likely the ones that are unions of paths are most interesting, since we want there to be many paths before we set weights.
'''
paths_as_edge_sets = [set(path_to_edge(path)) for path in paths]
merged = []
num_paths = len(paths_as_edge_sets)
if random == False:
path_iterator = itertools.combinations(paths_as_edge_sets, r)
else:
path_iterator = random_subset_iterator(paths_as_edge_sets, r)
for path_collection in path_iterator:
merge = set()
for path in path_collection:
merge = merge | path
subgraph = nx.DiGraph(nx.edge_subgraph(graph, list(merge)))
for x in graph.nodes():
if x not in subgraph.nodes():
subgraph.add_node(x)
t = 0
for x in subgraph.nodes():
#print(x)
subgraph.nodes[x]["pos"] = np.array([x[0], x[1]])
subgraph.nodes[x]["label"] = t
t += 1
subgraph.nodes[x]["weight"] = 0
yield subgraph
def node(node_id):
return str(
json.dumps(
nx.node_link_data(
nx.edge_subgraph(G,
list(nx.edges(G, node_id))[:20]))))
def test_edge_subgraph(self):
assert_equal(self.G.edge_subgraph([(1, 2), (0, 3)]).adj,
nx.edge_subgraph(self.G, [(1, 2), (0, 3)]).adj)
assert_equal(self.DG.edge_subgraph([(1, 2), (0, 3)]).adj,
nx.edge_subgraph(self.DG, [(1, 2), (0, 3)]).adj)
def graph(self):
return nx.edge_subgraph(self._graph, self.reference_edges)
def statistics():
trials = 100
#samples = trials*500
m = 20
graph = nx.grid_graph([m, m])
graph.name = "grid_size:" + str(m)
print(graph.name)
for x in graph.nodes():
graph.nodes[x]["pos"] = np.array([x[0], x[1]])
dual = Facefinder.planar_dual(graph)
W_trees = []
branches = []
supernode = find_supernode(dual)
boundary_faces = list(dual.neighbors(supernode))
face_1 = boundary_faces[0]
face_2 = boundary_faces[int(len(boundary_faces) / 2) + 1]
#if (face_1, supernode) or (supernode, face_1) in dual.edges():
# print("OK")
#if (face_2, supernode) or (supernode, face_2) in dual.edges():
# print("OK2")
cycles = []
cycles_containing_prescribed_faces = []
corresponding_walks = []
boundary_faces_frequencies = {}
for face_a in boundary_faces:
for face_b in boundary_faces:
boundary_faces_frequencies[(face_a, face_b)] = 0
#print("testing", boundary_faces_frequencies[ ( face_1, face_2)])
done = False
sample_counter = 0
while not done:
tree = nx.to_undirected(random_spanning_tree_wilson(dual))
available_edges = set(dual.edges())
available_edges = available_edges - set(tree.edges())
e = random.choice(list(available_edges))
unicycle = copy.deepcopy(tree)
unicycle = nx.Graph(unicycle)
unicycle.add_edge(e[0], e[1])
#cycles.append(simple_cycle(unicycle))
cycle = simple_cycle(unicycle)
#faces = [x[0] for x in cycle] + [x[1] for x in cycle]
#faces = set(faces)
#print(faces)
for face_a in boundary_faces:
for face_b in boundary_faces:
if (face_a, supernode) in cycle or (supernode,
face_a) in cycle:
if (face_b, supernode) in cycle or (supernode,
face_b) in cycle:
boundary_faces_frequencies[(face_a, face_b)] += 1
face_a = face_1
face_b = face_2
if (face_a, supernode) in cycle or (supernode, face_a) in cycle:
if (face_b, supernode) in cycle or (supernode, face_b) in cycle:
#print('1')
cycles_containing_prescribed_faces.append(cycle)
walk = nx.Graph(nx.edge_subgraph(dual, cycle))
walk.remove_node(supernode)
corresponding_walks.append(walk)
sample_counter += 1
print(sample_counter)
if sample_counter == trials:
done = True
#print("testing2", boundary_faces_frequencies[ (face_1, face_2)])
#print(corresponding_walks[0].edges())
#print(len(cycles_containing_prescribed_faces))
#print(boundary_faces_frequencies)
print("finished with cycle portions")
LERW = []
dual.remove_node(supernode)
#Because we are testing whether the distributions looks like a LERW in the grid
#portion of the dual graph
for i in range(trials):
trip = random_walk_until_hit(dual, face_1, set([face_2]))
new_branch, branch_length = loop_erasure(trip)
LERW.append(new_branch)
return LERW, corresponding_walks, dual, boundary_faces_frequencies
def to_edge_type_graph(self, edge_type):
type_edges = self.edges[self.edges["edge_type"] == edge_type]
view = nx.edge_subgraph(self.g, type_edges.index)
return MaggotGraph(view, self.nodes, type_edges)
def filtered_graph(graph, min_clearing):
subG = nx.edge_subgraph(
graph,
((u, v)
for u, v, c in graph.edges(data='clearing') if c >= min_clearing))
return subG # max(nx.connected_component_subgraphs(subG), key=len)
def edge_subgraph(self, edges):
return nx.edge_subgraph(self.Graph, edges)
generation = newgen.copy()
prevdist = bestdist
try:
'''
Draw visual representation
'''
# Plot matrix path
if matrix_plot:
'''
Adjacency Matrix plot
'''
matrix = np.array(mat)
g = nx.from_numpy_matrix(matrix)
g = nx.relabel_nodes(g, dict(zip(range(len(labels)), labels)))
g = nx.edge_subgraph(g, [(labels[best[i]], labels[best[i + 1]])
for i in range(len(best) - 1)])
pos = nx.shell_layout(g)
'''
Path
'''
nx.draw_networkx(g)
plt.title("Traveling Salesman Route")
plt.show()
# Plot points and connections
elif points_plot:
x = [points[best[i]][0] for i in range(len(points))]
y = [points[best[i]][1] for i in range(len(points))]
for i in range(len(best) - 1):
plt.plot(x[i:i + 2], y[i:i + 2], 'ro-')
plt.show()
except:
def subgraph(self, edges):
return nx.DiGraph(nx.edge_subgraph(self.graph(), edges))