def to_directed(self, as_view=False):
"""Return a directed representation of the graph.
Returns
-------
G : MultiDiGraph
A directed graph with the same name, same nodes, and with
each edge (u, v, data) replaced by two directed edges
(u, v, data) and (v, u, data).
Notes
-----
This returns a "deepcopy" of the edge, node, and
graph attributes which attempts to completely copy
all of the data and references.
This is in contrast to the similar D=DiGraph(G) which returns a
shallow copy of the data.
See the Python copy module for more information on shallow
and deep copies, https://docs.python.org/2/library/copy.html.
Warning: If you have subclassed MultiGraph to use dict-like objects
in the data structure, those changes do not transfer to the
MultiDiGraph created by this method.
Examples
--------
>>> G = nx.Graph() # or MultiGraph, etc
>>> G.add_edge(0, 1)
>>> H = G.to_directed()
>>> list(H.edges)
[(0, 1), (1, 0)]
If already directed, return a (deep) copy
>>> G = nx.DiGraph() # or MultiDiGraph, etc
>>> G.add_edge(0, 1)
>>> H = G.to_directed()
>>> list(H.edges)
[(0, 1)]
"""
if as_view is True:
return nx.graphviews.MultiDiGraphView(self)
# deepcopy when not a view
from networkx.classes.multidigraph import MultiDiGraph
G = MultiDiGraph()
G.graph.update(deepcopy(self.graph))
G.add_nodes_from((n, deepcopy(d)) for n, d in self._node.items())
G.add_edges_from((u, v, key, deepcopy(datadict))
for u, nbrs in self.adj.items()
for v, keydict in nbrs.items()
for key, datadict in keydict.items())
return G
def add_node(self, n,**kwds):
if n not in self:
MultiDiGraph.add_node(self,n)
self.nodeid[n]=self.nextid
self.nextid+=1
self.ubigraph.new_vertex_w_id(self.nodeid[n])
# add ubigraph attributes
for (k,v) in kwds.items():
ret=self.ubigraph.set_vertex_attribute(self.nodeid[n],k,v)
# support toggling node labels
if self.use_node_labels:
self.ubigraph.set_vertex_attribute(self.nodeid[n],'label',str(n))
def add_node(self, n, **kwds):
if n not in self:
MultiDiGraph.add_node(self, n)
self.nodeid[n] = self.nextid
self.nextid += 1
self.ubigraph.new_vertex_w_id(self.nodeid[n])
# add ubigraph attributes
for (k, v) in kwds.items():
ret = self.ubigraph.set_vertex_attribute(self.nodeid[n], k, v)
# support toggling node labels
if self.use_node_labels:
self.ubigraph.set_vertex_attribute(self.nodeid[n], 'label',
str(n))
def add_snapshot(self,
ebunch=None,
graph=None,
start=None,
end=None,
time=None,
multi=False):
"""Add a snapshot with a bunch of edge values.
Parameters
----------
ebunch : container of edges, optional (default= None)
Each edge in the ebunch list will be included to all added graphs.
graph : networkx graph object, optional (default= None)
networkx graph to be inserted into snapshot graph.
start: start timestamp, inclusive
end: end timestamp, exclusive
time: timestamp for impulses, cannot be used together with (start, end)
multi : boolean, optional (default= False)
Determines if type of graphs in snapshot are DiGraphs for MultiDiGraphs
Returns
-------
None
Examples
--------
>>> G = dnx.SnapshotGraph()
>>> G.add_snapshot([(1, 4), (1, 3)])
"""
if not graph:
if multi is True:
g = MultiDiGraph()
else:
g = DiGraph()
g.add_edges_from(ebunch)
else:
g = graph
if time is not None and (start or end):
raise ValueError(
'Time and (start or end) cannot both be specified.')
elif time is not None:
self.insert(g, time=time)
elif (start is None and end is not None) or (start is not None
and end is None):
raise ValueError(
'Start and end must both be specified for intervals.')
else:
self.insert(g, start=start, end=end)
def to_directed(self):
"""Return a directed representation of the graph.
Returns
-------
G : MultiDiGraph
A directed graph with the same name, same nodes, and with
each edge (u,v,data) replaced by two directed edges
(u,v,data) and (v,u,data).
Notes
-----
This returns a "deepcopy" of the edge, node, and
graph attributes which attempts to completely copy
all of the data and references.
This is in contrast to the similar D=DiGraph(G) which returns a
shallow copy of the data.
See the Python copy module for more information on shallow
and deep copies, http://docs.python.org/library/copy.html.
Warning
-------
If you have subclassed MultiGraph to use dict-like objects in the
data structure, those changes do not transfer to the MultiDiGraph
created by this method.
Examples
--------
>>> G = nx.Graph() # or MultiGraph, etc
>>> G.add_path([0,1])
>>> H = G.to_directed()
>>> H.edges()
[(0, 1), (1, 0)]
If already directed, return a (deep) copy
>>> G = nx.DiGraph() # or MultiDiGraph, etc
>>> G.add_path([0,1])
>>> H = G.to_directed()
>>> H.edges()
[(0, 1)]
"""
from networkx.classes.multidigraph import MultiDiGraph
G=MultiDiGraph()
G.add_nodes_from(self)
G.add_edges_from( (u,v,key,deepcopy(datadict))
for u,nbrs in self.adjacency_iter()
for v,keydict in nbrs.items()
for key,datadict in keydict.items() )
G.graph=deepcopy(self.graph)
G.node=deepcopy(self.node)
return G
def plot_vehicle_in_graph(graph: MultiDiGraph, vehicles: List[Vehicle]):
"""
在路网图上绘制车辆
:param graph:
:param vehicles:
:return:
"""
vehicle_nodes = set()
for vehicle in vehicles:
vehicle_nodes.add(GeoLocation.INDEX2OSM_ID[vehicle.location.osm_index])
nc = ['r' if node in vehicle_nodes else 'none' for node in graph.nodes()]
ox.plot_graph(graph, node_color=nc)
def to_graphics(A):
G = MultiDiGraph()
G.add_nodes_from(A.vertices())
for u, v in A.edges():
for a, b, d in A.weights(u, v):
G.add_edge(u, v, label=a + "->" + b + "," + d)
return G
def to_directed(self):
"""Return a directed representation of the nxgraph.
Returns
-------
G : MultiDiGraph
A directed nxgraph with the same name, same nodes, and with
each edge (u,v,data) replaced by two directed edges
(u,v,data) and (v,u,data).
Notes
-----
This returns a "deepcopy" of the edge, node, and
nxgraph attributes which attempts to completely copy
all of the data and references.
This is in contrast to the similar D=DiGraph(G) which returns a
shallow copy of the data.
See the Python copy module for more information on shallow
and deep copies, http://docs.python.org/library/copy.html.
Examples
--------
>>> G = nx.Graph() # or MultiGraph, etc
>>> G.add_path([0,1])
>>> H = G.to_directed()
>>> H.edges()
[(0, 1), (1, 0)]
If already directed, return a (deep) copy
>>> G = nx.DiGraph() # or MultiDiGraph, etc
>>> G.add_path([0,1])
>>> H = G.to_directed()
>>> H.edges()
[(0, 1)]
"""
from networkx.classes.multidigraph import MultiDiGraph
G = MultiDiGraph()
G.add_nodes_from(self)
G.add_edges_from(
(u, v, key, deepcopy(datadict))
for u, nbrs in self.adjacency_iter()
for v, keydict in nbrs.items()
for key, datadict in keydict.items()
)
G.graph = deepcopy(self.graph)
G.node = deepcopy(self.node)
return G
def read_graph(f, cfgOnly=False):
try:
if (f[-6:] == "pickle"):
return pickle.load(open(f, "rb"))
else:
G = MultiDiGraph(read_graphml(f))
if (cfgOnly or G.graph.get("nature", None) == "CFG"):
print("(removing PDG data from CFG)...")
G = _removePDGStructures(G)
else:
print("(fixing graph)...")
G = fix_graph(G)
#end if
return G
#end if
except Exception as e:
print >> stderr, ("ERROR: could not read graphML file " + f + "!")
print >> stderr, (str(e))
exit(1)
def __init__(self, G):
self.__graph = MultiDiGraph();
nodesToKeep = {n for n in G.nodes_iter(False) if is_cfg_node(G, n)};
self.__graph.add_nodes_from(nodesToKeep);
# mark the entry node
(self.__entryNode, isInterprocedural) = findGraphEntry(G);
assert(self.__entryNode in self.__graph);
# then, copy all CFG edges
for n in self.__graph.nodes_iter(False):
if(isInterprocedural and G.node[n].get("kind", "") == "call-site"):
foundOne = False;
for (source, target, attr) in G.out_edges_iter([n], data=True):
if(attr.get("type", "flow") == "control" and \
attr.get("scope", "") == "interprocedural"):
self.__graph.add_edge(n, target);
foundOne = True;
#end if
#end for
# add appropriate intraprocedural edges: only if
# (a) the called function is not in the graphml, or
# (b) the target is a crash point (which is essentially ambiguity
# nonsensemeaning that we crashed trying to make the call itself)
for (source, target, attr) in G.out_edges_iter([n], data=True):
if(attr.get("type", "flow") == "flow" and \
attr.get("scope", "") != "interprocedural" and \
(not foundOne or G.node[target].get("kind", "") == "crash")):
self.__graph.add_edge(n, target);
#end if
#end for
elif(isInterprocedural and G.node[n].get("kind", "") == "exit"):
entryForExit = findEntryForNode(G, n);
for (source, target, attr) in G.in_edges_iter([entryForExit], data=True):
if(attr.get("type", "flow") == "control" and \
attr.get("scope", "") == "interprocedural"):
# edge from exit -> all successors of the call to this function
for (call, callTarget, attr) in G.out_edges_iter([source], data=True):
if(attr.get("type", "flow") == "flow" and \
attr.get("scope", "") != "interprocedural" and \
G.node[callTarget].get("kind", "") != "crash"):
self.__graph.add_edge(n, callTarget);
#end if
#end for
#end if
#end for
else:
for (source, target, attr) in G.out_edges_iter([n], data=True):
if(attr.get("type", "flow") == "flow" and \
attr.get("scope", "") != "interprocedural"):
self.__graph.add_edge(n, target);
#end if
#end for
#end if
#end for
# setup for yes, no, maybe, and crash constraints
self.__crashNode = None;
self.__yesVectors = set([]);
self.__allYes = set([]);
self.__allNo = set([]);
def __init__(self, cx=None, server=None, username=None, password=None, uuid=None, data=None, **attr):
'''There are generally four ways to create a graph.
1. An empty graph. G = NdexGraph()
2. Using a cx dictionary. G = NdexGraph(cx)
3. Loading it from an NDEx server.
G = NdexGraph(server='http://test.ndexbio.org' uuid='983a2b93-2c55-11e6-a7c5-0630eb0972a1')
4. Just like any other NetworkX MultiDiGraph().
'''
MultiDiGraph.__init__(self, data, **attr)
self.subnetwork_id = None
self.view_id = None
self.max_node_id = None
self.max_edge_id = None
self.pos = {}
self.unknown_cx = []
# Maps edge ids to node ids. e.g. { edge1: (source_node, target_node), edge2: (source_node, target_node) }
self.edgemap = {}
if not cx and server and uuid:
ndex = nc.Ndex(server,username,password)
cx = ndex.get_network_as_cx_stream(uuid).json()
if not cx:
raise RuntimeError("Failed to retrieve network with uuid " + uuid + " from " + server)
# If there is no CX to process, just return.
if cx == None:
return
# First pass, get information about subnetworks.
for aspect in cx:
if 'subNetworks' in aspect:
for subnetwork in aspect['subNetworks']:
id = subnetwork['@id']
if self.subnetwork_id != None:
raise ValueError("networkn does not support collections!")
self.subnetwork_id = id
elif 'cyViews' in aspect:
for cyViews in aspect['cyViews']:
id = cyViews['@id']
if self.view_id != None:
raise ValueError("networkn does not support more than one view!")
self.view_id = id
elif 'metaData' in aspect:
# Strip metaData
continue
else:
self.unknown_cx.append(aspect)
cx = self.unknown_cx
# Second pass, just build basic graph.
self.unknown_cx = []
for aspect in cx:
if 'nodes' in aspect:
for node in aspect['nodes']:
id = node['@id']
name = node['n'] if 'n' in node else None
if name:
self.add_node(id, name=name)
else:
self.add_node(id)
represents = node['r'] if 'r' in node else None
if represents:
self.node[id]['represents'] = represents
elif 'edges' in aspect:
for edge in aspect['edges']:
id = edge['@id']
interaction = edge['i'] if 'i' in edge else None
s = edge['s']
t = edge['t']
self.edgemap[id] = (s, t)
if interaction:
self.add_edge(s, t, key=id, interaction=interaction)
else:
self.add_edge(s, t, key=id)
else:
self.unknown_cx.append(aspect)
cx = self.unknown_cx
# Third pass, handle attributes
# Notes. Not handled, datatypes.
self.unknown_cx = []
for aspect in cx:
if 'networkAttributes' in aspect:
for networkAttribute in aspect['networkAttributes']:
name = networkAttribute['n']
# special: ignore selected
if name == 'selected':
continue
value = networkAttribute['v']
if 'd' in networkAttribute:
d = networkAttribute['d']
if d == 'boolean':
value = value.lower() == 'true'
if 's' in networkAttribute or name not in self.graph:
self.graph[name] = value
elif 'nodeAttributes' in aspect:
for nodeAttribute in aspect['nodeAttributes']:
id = nodeAttribute['po']
name = nodeAttribute['n']
# special: ignore selected
if name == 'selected':
continue
value = nodeAttribute['v']
if 'd' in nodeAttribute:
d = nodeAttribute['d']
if d == 'boolean':
value = value.lower() == 'true'
if 's' in nodeAttribute or name not in self.node[id]:
self.node[id][name] = value
elif 'edgeAttributes' in aspect:
for edgeAttribute in aspect['edgeAttributes']:
id = edgeAttribute['po']
s, t = self.edgemap[id]
name = edgeAttribute['n']
# special: ignore selected and shared_name columns
if name == 'selected' or name == 'shared name':
continue
value = edgeAttribute['v']
if 'd' in edgeAttribute:
d = edgeAttribute['d']
if d == 'boolean':
value = value.lower() == 'true'
if 's' in edgeAttribute or name not in self[s][t][id]:
self[s][t][id][name] = value
else:
self.unknown_cx.append(aspect)
cx = self.unknown_cx
# Fourth pass, node locations
self.pos = {}
self.unknown_cx = []
for aspect in cx:
if 'cartesianLayout' in aspect:
for nodeLayout in aspect['cartesianLayout']:
id = nodeLayout['node']
x = nodeLayout['x']
y = nodeLayout['y']
self.pos[id] = np.array([x,y], dtype='float32')
else:
self.unknown_cx.append(aspect)
def delete_node(self, n):
if n in self:
MultiDiGraph.delete_node(self, n)
self.ubigraph.remove_vertex(self.nodeid[n])
def clear(self):
if len(self) > 0:
MultiDiGraph.clear(self)
self.ubigraph.clear()
self.nodeid = {}
self.nextid = 0
def to_subgraph(self,
begin,
end,
inclusive=(True, False),
multigraph=False,
edge_data=False,
edge_timestamp_data=False,
node_data=False):
"""Return a networkx Graph or MultiGraph which includes all the nodes and
edges which have timestamps within the given interval.
Parameters
----------
begin: int or float
end: int or float
Must be bigger than or equal to begin.
inclusive: 2-tuple boolean that determines inclusivity of begin and end
multigraph: bool, optional (default= False)
If True, a networkx MultiGraph will be returned. If False, networkx Graph.
edge_data: bool, optional (default= False)
If True, edges will keep their attributes.
edge_timestamp_data: bool, optional (default= False)
If True, each edge's attribute will also include its timestamp data.
If `edge_data= True` and there already exist edge attributes named timestamp
it will be overwritten.
node_data : bool, optional (default= False)
if True, each node's attributes will be included.
See Also
--------
to_snapshots : divide the impulse graph to snapshots
Notes
-----
If multigraph= False, and edge_data=True or edge_interval_data=True,
in case there are multiple edges, only one will show with one of the edge's attributes.
Note: nodes with no edges will not appear in any subgraph.
Examples
--------
>>> G = dnx.ImpulseGraph()
>>> G.add_edges_from([(1, 2, 10), (2, 4, 11), (6, 4, 19), (2, 4, 15)])
>>> H = G.to_subgraph(4, 12)
>>> type(H)
<class 'networkx.classes.graph.DiGraph'>
>>> list(H.edges(data=True))
[(1, 2, {}), (2, 4, {})]
>>> H = G.to_subgraph(10, 12, edge_timestamp_data=True)
>>> type(H)
<class 'networkx.classes.graph.DiGraph'>
>>> list(H.edges(data=True))
[(1, 2, {'timestamp': 10}), (2, 4, {'timestamp': 11})]
>>> M = G.to_subgraph(4, 12, multigraph=True, edge_timestamp_data=True)
>>> type(M)
<class 'networkx.classes.multigraph.MultiDiGraph'>
>>> list(M.edges(data=True))
[(1, 2, {'timestamp': 10}), (2, 4, {'timestamp': 11})]
"""
iedges = self.__search_tree(begin, end, inclusive=inclusive)
if multigraph:
G = MultiDiGraph()
else:
G = DiGraph()
if edge_data and edge_timestamp_data:
G.add_edges_from((iedge[0], iedge[1],
dict(self._pred[iedge[0]][iedge[1]][iedge],
timestamp=iedge[3])) for iedge in iedges)
elif edge_data:
G.add_edges_from(
(iedge[0], iedge[1], self._pred[iedge[0]][iedge[1]][iedge])
for iedge in iedges)
elif edge_timestamp_data:
G.add_edges_from((iedge[0], iedge[1], {
'timestamp': iedge[3]
}) for iedge in iedges)
else:
G.add_edges_from((iedge[0], iedge[1]) for iedge in iedges)
if node_data:
G.add_nodes_from((n, self._node[n].copy()) for n in G.nodes)
return G
def to_subgraph(self,
begin,
end,
multigraph=False,
edge_data=False,
edge_interval_data=False,
node_data=False):
"""Return a networkx DiGraph or MultiDiGraph which includes all the nodes and
edges which have overlapping intervals with the given interval.
Parameters
----------
begin: int or float
Inclusive beginning time of the edge appearing in the interval graph.
end: int or float
Non-inclusive ending time of the edge appearing in the interval graph.
Must be bigger than or equal to begin.
multigraph: bool, optional (default= False)
If True, a networkx MultiGraph will be returned. If False, networkx Graph.
edge_data: bool, optional (default= False)
If True, edges will keep their attributes.
edge_interval_data: bool, optional (default= False)
If True, each edge's attribute will also include its begin and end interval data.
If `edge_data= True` and there already exist edge attributes with names begin and end,
they will be overwritten.
node_data : bool, optional (default= False)
If True, each node's attributes will be included.
See Also
--------
to_snapshots : divide the interval graph to snapshots
Notes
-----
If multigraph= False, and edge_data=True or edge_interval_data=True,
in case there are multiple edges, only one will show with one of the edge's attributes.
Note: nodes with no edges will not appear in any subgraph.
Examples
--------
>>> G = dnx.IntervalGraph()
>>> G.add_edges_from([(1, 2, 3, 10), (2, 4, 1, 11), (6, 4, 12, 19), (2, 4, 8, 15)])
>>> H = G.to_subgraph(4, 12)
>>> type(H)
<class 'networkx.classes.graph.DiGraph'>
>>> list(H.edges(data=True))
[(1, 2, {}), (2, 4, {})]
>>> H = G.to_subgraph(4, 12, edge_interval_data=True)
>>> type(H)
<class 'networkx.classes.graph.DiGraph'>
>>> list(H.edges(data=True))
[(1, 2, {'end': 10, 'begin': 3}), (2, 4, {'end': 15, 'begin': 8})]
>>> M = G.to_subgraph(4, 12, multigraph=True, edge_interval_data=True)
>>> type(M)
<class 'networkx.classes.multigraph.MultiDiGraph'>
>>> list(M.edges(data=True))
[(1, 2, {'end': 10, 'begin': 3}), (2, 4, {'end': 11, 'begin': 1}), (2, 4, {'end': 15, 'begin': 8})]
"""
if begin and end and begin > end:
raise NetworkXError(
"IntervalGraph: interval end must be bigger than or equal to begin: "
"begin: {}, end: {}.".format(begin, end))
iedges = self.tree[begin:end]
if multigraph:
G = MultiDiGraph()
else:
G = DiGraph()
if edge_data and edge_interval_data:
G.add_edges_from((iedge.data[0], iedge.data[1],
dict(self._adj[iedge.data[0]][iedge],
begin=iedge.begin,
end=iedge.end)) for iedge in iedges)
elif edge_data:
G.add_edges_from((iedge.data[0], iedge.data[1],
self._adj[iedge.data[0]][iedge].copy())
for iedge in iedges)
elif edge_interval_data:
G.add_edges_from((iedge.data[0], iedge.data[1], {
'begin': iedge.begin,
'end': iedge.end
}) for iedge in iedges)
else:
G.add_edges_from(
(iedge.data[0], iedge.data[1]) for iedge in iedges)
# include node attributes
if node_data:
G.add_nodes_from((n, self._node[n].copy()) for n in G.nodes)
return G
def delete_node(self,n):
if n in self:
MultiDiGraph.delete_node(self,n)
self.ubigraph.remove_vertex(self.nodeid[n])
class UtlExecutionSolver(ExecutionSolver):
__slots__ = "__graph, __entryNode, __crashNode, __yesVectors, __allYes, __allNo";
"""
@override
__init__(): Process the graph, making our own CFG-only, massaged copy.
@param G the graph
"""
def __init__(self, G):
self.__graph = MultiDiGraph();
nodesToKeep = {n for n in G.nodes_iter(False) if is_cfg_node(G, n)};
self.__graph.add_nodes_from(nodesToKeep);
# mark the entry node
(self.__entryNode, isInterprocedural) = findGraphEntry(G);
assert(self.__entryNode in self.__graph);
# then, copy all CFG edges
for n in self.__graph.nodes_iter(False):
if(isInterprocedural and G.node[n].get("kind", "") == "call-site"):
foundOne = False;
for (source, target, attr) in G.out_edges_iter([n], data=True):
if(attr.get("type", "flow") == "control" and \
attr.get("scope", "") == "interprocedural"):
self.__graph.add_edge(n, target);
foundOne = True;
#end if
#end for
# add appropriate intraprocedural edges: only if
# (a) the called function is not in the graphml, or
# (b) the target is a crash point (which is essentially ambiguity
# nonsensemeaning that we crashed trying to make the call itself)
for (source, target, attr) in G.out_edges_iter([n], data=True):
if(attr.get("type", "flow") == "flow" and \
attr.get("scope", "") != "interprocedural" and \
(not foundOne or G.node[target].get("kind", "") == "crash")):
self.__graph.add_edge(n, target);
#end if
#end for
elif(isInterprocedural and G.node[n].get("kind", "") == "exit"):
entryForExit = findEntryForNode(G, n);
for (source, target, attr) in G.in_edges_iter([entryForExit], data=True):
if(attr.get("type", "flow") == "control" and \
attr.get("scope", "") == "interprocedural"):
# edge from exit -> all successors of the call to this function
for (call, callTarget, attr) in G.out_edges_iter([source], data=True):
if(attr.get("type", "flow") == "flow" and \
attr.get("scope", "") != "interprocedural" and \
G.node[callTarget].get("kind", "") != "crash"):
self.__graph.add_edge(n, callTarget);
#end if
#end for
#end if
#end for
else:
for (source, target, attr) in G.out_edges_iter([n], data=True):
if(attr.get("type", "flow") == "flow" and \
attr.get("scope", "") != "interprocedural"):
self.__graph.add_edge(n, target);
#end if
#end for
#end if
#end for
# setup for yes, no, maybe, and crash constraints
self.__crashNode = None;
self.__yesVectors = set([]);
self.__allYes = set([]);
self.__allNo = set([]);
#end: __init__
"""
__findSCCFromNode(): Find the SCC in G that contains n.
@param G the graph
@param nodeToFind the node to find
@return the node in G that contains n within its SCC (throws an exception if
not found)
"""
def __findSCCFromNode(self, G, nodeToFind):
for (n, data) in G.nodes_iter(data=True):
if(nodeToFind in data.get("members", [])):
return(n);
#end for
raise KeyError("Node " + str(nodeToFind) + " not found.");
#end: __findSCCFromNode
"""
__isDAG(): Verify that the graph is a DAG (i.e., has no cycles).
@param G the graph
@return true if G is a DAG, otherwise false
"""
def __isDAG(self, G):
# TODO: efficiency (this could be O(n))
for n in G.nodes_iter(data=False):
visited = set([]);
worklist = deque([target for (source, target) in G.out_edges_iter([n])]);
while(worklist):
current = worklist.popleft();
if(current == n):
return(False);
elif(current in visited):
continue;
else:
visited.add(current);
#end if
for (source, target) in G.out_edges_iter([current]):
worklist.append(target);
#end for
#end while
#end for
return(True);
#end: __isDAG
"""
__clearNodeFacts(): Clear old before- and after-facts from all nodes in the
SCC graph.
@param G the graph
NOTE: modifies G in-place
"""
def __clearNodeFacts(self, G):
for (n, data) in G.nodes_iter(data=True):
data["afterFact"] = None;
data["beforeFact"] = None;
#end for
#end: __clearNodeFacts
"""
__isPrefix(): Check if the first list is a prefix of the second.
@param first the first list
@param second the second list
@return true if first is a prefix of second, false otherwise
"""
def __isPrefix(self, first, second):
if(len(second) < len(first)):
return(False);
#end if
for (i, val) in enumerate(first):
if(second[i] != val):
return(False);
#end if
#end for
return(True);
#end: __isPrefix
"""
__getSCCEntryCrash(): Get the SCC for the entry and the crash node. It is an
error if either is not found because G is not a SCC DAG.
@param G the graph
@return a tuple: (entrySCC, crashSCC)
"""
def __getSCCEntryCrash(self, G):
sccEntry = self.__findSCCFromNode(G, self.__entryNode);
sccCrash = self.__findSCCFromNode(G, self.__crashNode);
return(sccEntry, sccCrash);
#end: __getSCCEntryCrash
"""
__backwardReachableFrom(): Get all nodes in G backward-reachable from n.
@param G the graph
@param n the node to reach backward from
@return the set of nodes (including n itself)
"""
def __backwardReachableFrom(self, G, n):
reachable = set([]);
worklist = [n];
while(worklist):
current = worklist.pop();
if(current in reachable):
continue;
reachable.add(current);
for (source, target) in G.in_edges_iter([current]):
worklist.append(source);
#end for
#end while
return(reachable);
#end: __backwardReachableFrom
"""
__recReverseTopo(): A recursive sub-procedure for __reverseTopoOrdering().
@param G the graph, must be a DAG (i.e., SCC collapsed)
@param n the node (from G) to recurse backward from
@param ordering a deque that will hold the final ordering
@param temporaryMark internal marking for "in-progress" nodes
@param doneMark internal marking for completely processed nodes
"""
def __recReverseTopo(self, G, n, ordering, temporaryMark, doneMark):
if(n in doneMark):
return;
elif(n in temporaryMark):
print >> stderr, ("ERROR: graph for rev topo is not a DAG!");
exit(1);
#end if
temporaryMark.add(n);
for (source, target) in G.in_edges_iter([n]):
self.__recReverseTopo(G, source, ordering, temporaryMark, doneMark);
#end for
assert(n not in doneMark);
doneMark.add(n);
ordering.appendleft(n);
#end: __recReverseTopo
"""
__reverseTopoOrdering(): Get a reverse topological ordering of the nodes in
G, including only those nodes backward-reachable from
crash. Note that G must be an acyclic graph.
Further, this ordering may not include all of G's
nodes, meaning that some nodes WON'T have all
decendants processed before themselves...but they
will only exclude those nodes that are not
backward-reachable from crash.
@param G the graph, must be a DAG (i.e., SCC collapsed)
@param crash the ending point to use for the reverse ordering
@return a reverse topological ordering of G's nodes, starting with crash.
Crashes with error if G is not a DAG.
"""
def __reverseTopoOrdering(self, G, crash):
ordering = deque();
temporaryMark = set([]);
doneMark = set([]);
self.__recReverseTopo(G, crash, ordering, temporaryMark, doneMark);
return(ordering);
#end: __reverseTopoOrdering
"""
__entryCrashPath(): Check if there exists a path in the DAG from the
entry node's SCC, to the crash node's SCC. The path must
also meet all the obsYes (__yesVectors) conditions.
@param G the graph, must be a DAG (i.e., SCC collapsed)
@return true if a consistent path entry->crash exists, otherwise false
"""
def __entryCrashPath(self, G):
# an empty graph clearly has no path from entry->crash
if(not G):
return(False);
#end if
self.__clearNodeFacts(G);
# get the SCC for the entry and the crash
(realEntry, realCrash) = self.__getSCCEntryCrash(G);
# get those nodes backward-reachable from the crash
crashReachable = self.__backwardReachableFrom(G, realCrash);
# get a reverse topological order of nodes in G (up to and including the
# crash)
revTopoOrdering = self.__reverseTopoOrdering(G, realCrash);
assert(set(revTopoOrdering) == crashReachable);
for processing in revTopoOrdering:
# ---------------------------------------------------------------------
# compute this node's "after-fact" from the union of all incoming nodes
# (i.e., the incoming fact from children)
# ---------------------------------------------------------------------
afterFact = None;
for (source, target) in G.out_edges_iter([processing]):
childFact = G.node[target].get("beforeFact", None);
if(childFact == None):
assert(target not in crashReachable);
continue;
else:
assert(len(childFact) == len(self.__yesVectors));
#end if
if(afterFact == None):
afterFact = childFact;
continue;
else:
assert(len(childFact) == len(afterFact));
#end if
# the hard case: in order to be a consistent path, one child must have
# the smallest vector across all obsYes entries
newIsSmaller = False;
oldIsSmaller = False;
for (i, obsYesVector) in enumerate(childFact):
thisNewSmaller = self.__isPrefix(obsYesVector, afterFact[i]);
thisOldSmaller = self.__isPrefix(afterFact[i], obsYesVector);
assert(thisNewSmaller or thisOldSmaller);
# update which is the smaller vector (more obsYes entries eaten)
if(thisNewSmaller and thisOldSmaller):
# they are the same: always OK
continue;
elif(thisNewSmaller):
if(oldIsSmaller):
# KABOOM!
return(False);
#end if
newIsSmaller = True;
else:
if(newIsSmaller):
# KABOOM!
return(False);
#end if
oldIsSmaller = True;
#end if
#end for
assert(not newIsSmaller or not oldIsSmaller);
if(newIsSmaller):
afterFact = childFact;
#end if
#end for
# generate the "base" after-fact (for starting at the crash)
if(afterFact == None):
assert(processing == realCrash);
afterFact = [];
for obsYesVector in self.__yesVectors:
# NOTE: don't need to do a deep copy here, because we will copy and
# update the vector later (when computing the before-fact. Keep an
# eye on this for future changes!
afterFact += [list(obsYesVector)];
#end for
#end if
# ---------------------------------------------------------------------
# compute this node's "before-fact"
# ---------------------------------------------------------------------
beforeFact = [];
# build the new before-fact
for vector in afterFact:
# create our own copy of the fact, remove any nodes from our SCC from
# the end of the vector
newVector = vector[:];
while(newVector and newVector[-1] in G.node[processing]["members"]):
newVector.pop();
#end while
beforeFact.append(newVector);
#end for
# TODO: if any nodes from this SCC still in any vectors, KABOOM!
# update the before-fact
assert(G.node[processing].get("beforeFact", None) == None);
G.node[processing]["beforeFact"] = beforeFact;
#end for
# check the fact at the entry
entryBeforeFact = G.node[realEntry].get("beforeFact", None);
if(entryBeforeFact == None):
return(False);
assert(len(entryBeforeFact) == len(self.__yesVectors));
for vector in entryBeforeFact:
if(len(vector) > 0):
return(False);
#end if
#end for
return(True);
#end: __entryCrashPath
"""
__removeDeadNodes(): Remove all nodes that are either
(1) not forward-reachable from entry, or
(2) not backward-reachable from the crash site
@param G the graph (may be either a standard graph or an SCC DAG)
NOTE: G is updated in-place
"""
def __removeDeadNodes(self, G):
assert(self.__entryNode);
assert(self.__crashNode);
if(self.__entryNode in G and self.__crashNode in G):
entryNode = self.__entryNode;
crashNode = self.__crashNode;
else:
(entryNode, crashNode) = self.__getSCCEntryCrash(G);
#end if
fwdNodes = set([]);
worklist = set([entryNode]);
while(worklist):
current = worklist.pop();
fwdNodes.add(current);
for (source, target) in G.out_edges_iter([current]):
if(target not in fwdNodes):
worklist.add(target);
#end for
#end while
bwdNodes = set([]);
worklist = set([crashNode]);
while(worklist):
current = worklist.pop();
bwdNodes.add(current);
for (source, target) in G.in_edges_iter([current]):
if(source not in bwdNodes):
worklist.add(source);
#end for
#end while
oldNodes = set(G.nodes());
newNodes = oldNodes & fwdNodes & bwdNodes;
G.remove_nodes_from(oldNodes - newNodes);
#end: __removeDeadNodes
"""
__buildSCCGraph(): Build a new graph with SCCs from the input graph collapsed.
NOTE: the created graph contains only SCCs backward-reachable from the crash
node (which must be set before calling this function).
@param G the graph
@return the new SCC graph
"""
def __buildSCCGraph(self, G):
newGraph = condensation(G);
self.__removeDeadNodes(newGraph);
return(newGraph);
#end: __buildSCCGraph
"""
@override
isSat(): Check if there is a path from entry to crash.
@return whether or not any such path exists
"""
def isSat(self):
assert(self.__entryNode != None);
assert(self.__crashNode != None);
return(self.__entryCrashPath(self.__buildSCCGraph(self.__graph)));
#end: isSat
"""
@override
encodeObsYes(): Encode the constraint for a yes-executed observation.
@param possibleYes a set of possible matches to the true entry
(NOTE: currently only supports a singleton)
=> [{G.nodes}]
"""
def encodeObsYes(self, possibleYes):
thisEntry = [];
for group in possibleYes:
assert(len(group) == 1);
n = next(iter(group));
self.__allYes.add(n);
thisEntry.append(n);
#end for
self.__yesVectors.add(tuple(thisEntry));
#end: encodeObsYes
"""
@override
encodeCrash(): Encode the constraint for the crashing location.
@param crashStack a representation of possible crashes in the stack trace,
ending in the final possible crashing nodes
=> [({G.nodes}, {G.nodes}), ..., ({G.nodes}, None)]
"""
def encodeCrash(self, crashStack):
assert(self.__crashNode is None);
obsCrash = [];
for (callNodes, entryNodes) in crashStack:
obsCrash.append(callNodes);
if(entryNodes):
obsCrash.append(entryNodes);
#end for
self.encodeObsYes(obsCrash);
assert(len(crashStack[-1]) == 2);
crashNodes = crashStack[-1][0];
assert(len(crashNodes) == 1);
self.__crashNode = next(iter(crashNodes));
#end: encodeCrash
"""
@override
encodeObsNo(): Encode the constraint for a not-executed observation.
@param possibleNo a set of possible matches to the true entry
(NOTE: currently only supports a singleton)
=> {G.nodes}
"""
def encodeObsNo(self, possibleNo):
assert(len(possibleNo) == 1);
# remove node from the graph
n = next(iter(possibleNo));
self.__graph.remove_node(n);
self.__allNo.add(n);
# TODO: remove all now-disconnected nodes in the graph?
# it's probably better to just do this once, after all "no" observations
#end: encodeObsNo
"""
@override
findKnownExecution(): Figure out which nodes in the CFG (a) are known to have
executed at least once, (b) are known to have not executed, and (c) may or may
not have executed given the crash location.
@return (defYes, defNo, maybe)
=> ({G.nodes}, {G.nodes}, {G.nodes})
"""
def findKnownExecution(self):
defYes = set([]);
defNo = self.__allNo.copy();
maybe = set([]);
# build the base SCC graph (which is re-used for each exeNo check)
baseSCCGraph = self.__buildSCCGraph(self.__graph);
total = len(self.__graph);
soFar = 0;
for n in self.__graph.nodes_iter(False):
prevInYesVectors = (tuple([n]) in self.__yesVectors);
self.__yesVectors.add(tuple([n]));
possibleYes = self.__entryCrashPath(baseSCCGraph);
if(not prevInYesVectors):
self.__yesVectors.remove(tuple([n]));
#end if
if(n in [self.__entryNode, self.__crashNode]):
possibleNo = False;
else:
# make a shallow copy (so we can remove a node without wrecking the
# original)
noTestG = self.__graph.subgraph(self.__graph.nodes());
noTestG.remove_node(n);
noTestSCCGraph = self.__buildSCCGraph(noTestG);
possibleNo = self.__entryCrashPath(noTestSCCGraph);
#end if
if(not possibleYes and not possibleNo):
print >> stderr, ("ERROR: graph node " + n + " neither executed nor " +\
"didn't execute!");
exit(1);
elif(possibleYes and possibleNo):
maybe.add(n);
elif(possibleYes):
defYes.add(n);
else:
defNo.add(n);
soFar += 1;
if(soFar % 10 == 0):
stdout.write("\r" + ("%.2f" % ((1.0*soFar)/(1.0*total)*100)) + "%: " + \
str(soFar) + " / " + str(total));
stdout.flush();
#end if
#end for
print("");
return(defYes, defNo, maybe);
def clear(self):
if len(self)>0:
MultiDiGraph.clear(self)
self.ubigraph.clear()
self.nodeid={}
self.nextid=0
def __init__(self,
cx=None,
server=None,
username=None,
password=None,
uuid=None,
data=None,
**attr):
'''There are generally four ways to create a graph.
1. An empty graph. G = NdexGraph()
2. Using a cx dictionary. G = NdexGraph(cx)
3. Loading it from an NDEx server.
G = NdexGraph(server='http://test.ndexbio.org' uuid='983a2b93-2c55-11e6-a7c5-0630eb0972a1')
4. Just like any other NetworkX MultiDiGraph().
'''
MultiDiGraph.__init__(self, data, **attr)
self.subnetwork_id = None
self.view_id = None
self.max_node_id = None
self.max_edge_id = None
self.pos = {}
self.unknown_cx = []
# Maps edge ids to node ids. e.g. { edge1: (source_node, target_node), edge2: (source_node, target_node) }
self.edgemap = {}
if not cx and server and uuid:
ndex = nc.Ndex(server, username, password)
cx = ndex.get_network_as_cx_stream(uuid).json()
if not cx:
raise RuntimeError("Failed to retrieve network with uuid " +
uuid + " from " + server)
# If there is no CX to process, just return.
if cx == None:
return
# First pass, get information about subnetworks.
for aspect in cx:
if 'subNetworks' in aspect:
for subnetwork in aspect['subNetworks']:
id = subnetwork['@id']
if self.subnetwork_id != None:
raise ValueError(
"networkn does not support collections!")
self.subnetwork_id = id
elif 'cyViews' in aspect:
for cyViews in aspect['cyViews']:
id = cyViews['@id']
if self.view_id != None:
raise ValueError(
"networkn does not support more than one view!")
self.view_id = id
elif 'metaData' in aspect:
# Strip metaData
continue
else:
self.unknown_cx.append(aspect)
cx = self.unknown_cx
# Second pass, just build basic graph.
self.unknown_cx = []
for aspect in cx:
if 'nodes' in aspect:
for node in aspect['nodes']:
id = node['@id']
name = node['n'] if 'n' in node else None
if name:
self.add_node(id, name=name)
else:
self.add_node(id)
represents = node['r'] if 'r' in node else None
if represents:
self.node[id]['represents'] = represents
elif 'edges' in aspect:
for edge in aspect['edges']:
id = edge['@id']
interaction = edge['i'] if 'i' in edge else None
s = edge['s']
t = edge['t']
self.edgemap[id] = (s, t)
if interaction:
self.add_edge(s, t, key=id, interaction=interaction)
else:
self.add_edge(s, t, key=id)
else:
self.unknown_cx.append(aspect)
cx = self.unknown_cx
# Third pass, handle attributes
# Notes. Not handled, datatypes.
self.unknown_cx = []
for aspect in cx:
if 'networkAttributes' in aspect:
for networkAttribute in aspect['networkAttributes']:
name = networkAttribute['n']
# special: ignore selected
if name == 'selected':
continue
value = networkAttribute['v']
if 'd' in networkAttribute:
d = networkAttribute['d']
if d == 'boolean':
value = value.lower() == 'true'
if 's' in networkAttribute or name not in self.graph:
self.graph[name] = value
elif 'nodeAttributes' in aspect:
for nodeAttribute in aspect['nodeAttributes']:
id = nodeAttribute['po']
name = nodeAttribute['n']
# special: ignore selected
if name == 'selected':
continue
value = nodeAttribute['v']
if 'd' in nodeAttribute:
d = nodeAttribute['d']
if d == 'boolean':
value = value.lower() == 'true'
if 's' in nodeAttribute or name not in self.node[id]:
self.node[id][name] = value
elif 'edgeAttributes' in aspect:
for edgeAttribute in aspect['edgeAttributes']:
id = edgeAttribute['po']
s, t = self.edgemap[id]
name = edgeAttribute['n']
# special: ignore selected and shared_name columns
if name == 'selected' or name == 'shared name':
continue
value = edgeAttribute['v']
if 'd' in edgeAttribute:
d = edgeAttribute['d']
if d == 'boolean':
value = value.lower() == 'true'
if 's' in edgeAttribute or name not in self[s][t][id]:
self[s][t][id][name] = value
else:
self.unknown_cx.append(aspect)
cx = self.unknown_cx
# Fourth pass, node locations
self.pos = {}
self.unknown_cx = []
for aspect in cx:
if 'cartesianLayout' in aspect:
for nodeLayout in aspect['cartesianLayout']:
id = nodeLayout['node']
x = nodeLayout['x']
y = nodeLayout['y']
self.pos[id] = np.array([x, y], dtype='float32')
else:
self.unknown_cx.append(aspect)
