"""
Text-based visual representations of graphs
"""
import sys
import warnings
from collections import defaultdict
import networkx as nx
from networkx.utils import open_file
__all__ = ["generate_network_text", "write_network_text"]
class BaseGlyphs:
@classmethod
def as_dict(cls):
return {
a: getattr(cls, a)
for a in dir(cls)
if not a.startswith("_") and a != "as_dict"
}
class AsciiBaseGlyphs(BaseGlyphs):
empty: str = "+"
newtree_last: str = "+-- "
newtree_mid: str = "+-- "
endof_forest: str = " "
within_forest: str = ": "
within_tree: str = "| "
class AsciiDirectedGlyphs(AsciiBaseGlyphs):
last: str = "L-> "
mid: str = "|-> "
backedge: str = "<-"
vertical_edge: str = "!"
class AsciiUndirectedGlyphs(AsciiBaseGlyphs):
last: str = "L-- "
mid: str = "|-- "
backedge: str = "-"
vertical_edge: str = "|"
class UtfBaseGlyphs(BaseGlyphs):
# Notes on available box and arrow characters
# https://en.wikipedia.org/wiki/Box-drawing_character
# https://stackoverflow.com/questions/2701192/triangle-arrow
empty: str = "╙"
newtree_last: str = "╙── "
newtree_mid: str = "╟── "
endof_forest: str = " "
within_forest: str = "╎ "
within_tree: str = "│ "
class UtfDirectedGlyphs(UtfBaseGlyphs):
last: str = "└─╼ "
mid: str = "├─╼ "
backedge: str = "╾"
vertical_edge: str = "╽"
class UtfUndirectedGlyphs(UtfBaseGlyphs):
last: str = "└── "
mid: str = "├── "
backedge: str = "─"
vertical_edge: str = "│"
[docs]
def generate_network_text(
graph,
with_labels=True,
sources=None,
max_depth=None,
ascii_only=False,
vertical_chains=False,
):
"""Generate lines in the "network text" format
This works via a depth-first traversal of the graph and writing a line for
each unique node encountered. Non-tree edges are written to the right of
each node, and connection to a non-tree edge is indicated with an ellipsis.
This representation works best when the input graph is a forest, but any
graph can be represented.
This notation is original to networkx, although it is simple enough that it
may be known in existing literature. See #5602 for details. The procedure
is summarized as follows:
1. Given a set of source nodes (which can be specified, or automatically
discovered via finding the (strongly) connected components and choosing one
node with minimum degree from each), we traverse the graph in depth first
order.
2. Each reachable node will be printed exactly once on it's own line.
3. Edges are indicated in one of four ways:
a. a parent "L-style" connection on the upper left. This corresponds to
a traversal in the directed DFS tree.
b. a backref "<-style" connection shown directly on the right. For
directed graphs, these are drawn for any incoming edges to a node that
is not a parent edge. For undirected graphs, these are drawn for only
the non-parent edges that have already been represented (The edges that
have not been represented will be handled in the recursive case).
c. a child "L-style" connection on the lower right. Drawing of the
children are handled recursively.
d. if ``vertical_chains`` is true, and a parent node only has one child
a "vertical-style" edge is drawn between them.
4. The children of each node (wrt the directed DFS tree) are drawn
underneath and to the right of it. In the case that a child node has already
been drawn the connection is replaced with an ellipsis ("...") to indicate
that there is one or more connections represented elsewhere.
5. If a maximum depth is specified, an edge to nodes past this maximum
depth will be represented by an ellipsis.
6. If a node has a truthy "collapse" value, then we do not traverse past
that node.
Parameters
----------
graph : nx.DiGraph | nx.Graph
Graph to represent
with_labels : bool | str
If True will use the "label" attribute of a node to display if it
exists otherwise it will use the node value itself. If given as a
string, then that attribute name will be used instead of "label".
Defaults to True.
sources : List
Specifies which nodes to start traversal from. Note: nodes that are not
reachable from one of these sources may not be shown. If unspecified,
the minimal set of nodes needed to reach all others will be used.
max_depth : int | None
The maximum depth to traverse before stopping. Defaults to None.
ascii_only : Boolean
If True only ASCII characters are used to construct the visualization
vertical_chains : Boolean
If True, chains of nodes will be drawn vertically when possible.
Yields
------
str : a line of generated text
Examples
--------
>>> graph = nx.path_graph(10)
>>> graph.add_node("A")
>>> graph.add_node("B")
>>> graph.add_node("C")
>>> graph.add_node("D")
>>> graph.add_edge(9, "A")
>>> graph.add_edge(9, "B")
>>> graph.add_edge(9, "C")
>>> graph.add_edge("C", "D")
>>> graph.add_edge("C", "E")
>>> graph.add_edge("C", "F")
>>> nx.write_network_text(graph)
╙── 0
└── 1
└── 2
└── 3
└── 4
└── 5
└── 6
└── 7
└── 8
└── 9
├── A
├── B
└── C
├── D
├── E
└── F
>>> nx.write_network_text(graph, vertical_chains=True)
╙── 0
│
1
│
2
│
3
│
4
│
5
│
6
│
7
│
8
│
9
├── A
├── B
└── C
├── D
├── E
└── F
"""
from typing import Any, NamedTuple
class StackFrame(NamedTuple):
parent: Any
node: Any
indents: list
this_islast: bool
this_vertical: bool
collapse_attr = "collapse"
is_directed = graph.is_directed()
if is_directed:
glyphs = AsciiDirectedGlyphs if ascii_only else UtfDirectedGlyphs
succ = graph.succ
pred = graph.pred
else:
glyphs = AsciiUndirectedGlyphs if ascii_only else UtfUndirectedGlyphs
succ = graph.adj
pred = graph.adj
if isinstance(with_labels, str):
label_attr = with_labels
elif with_labels:
label_attr = "label"
else:
label_attr = None
if max_depth == 0:
yield glyphs.empty + " ..."
elif len(graph.nodes) == 0:
yield glyphs.empty
else:
# If the nodes to traverse are unspecified, find the minimal set of
# nodes that will reach the entire graph
if sources is None:
sources = _find_sources(graph)
# Populate the stack with each:
# 1. parent node in the DFS tree (or None for root nodes),
# 2. the current node in the DFS tree
# 2. a list of indentations indicating depth
# 3. a flag indicating if the node is the final one to be written.
# Reverse the stack so sources are popped in the correct order.
last_idx = len(sources) - 1
stack = [
StackFrame(None, node, [], (idx == last_idx), False)
for idx, node in enumerate(sources)
][::-1]
num_skipped_children = defaultdict(lambda: 0)
seen_nodes = set()
while stack:
parent, node, indents, this_islast, this_vertical = stack.pop()
if node is not Ellipsis:
skip = node in seen_nodes
if skip:
# Mark that we skipped a parent's child
num_skipped_children[parent] += 1
if this_islast:
# If we reached the last child of a parent, and we skipped
# any of that parents children, then we should emit an
# ellipsis at the end after this.
if num_skipped_children[parent] and parent is not None:
# Append the ellipsis to be emitted last
next_islast = True
try_frame = StackFrame(
node, Ellipsis, indents, next_islast, False
)
stack.append(try_frame)
# Redo this frame, but not as a last object
next_islast = False
try_frame = StackFrame(
parent, node, indents, next_islast, this_vertical
)
stack.append(try_frame)
continue
if skip:
continue
seen_nodes.add(node)
if not indents:
# Top level items (i.e. trees in the forest) get different
# glyphs to indicate they are not actually connected
if this_islast:
this_vertical = False
this_prefix = indents + [glyphs.newtree_last]
next_prefix = indents + [glyphs.endof_forest]
else:
this_prefix = indents + [glyphs.newtree_mid]
next_prefix = indents + [glyphs.within_forest]
else:
# Non-top-level items
if this_vertical:
this_prefix = indents
next_prefix = indents
else:
if this_islast:
this_prefix = indents + [glyphs.last]
next_prefix = indents + [glyphs.endof_forest]
else:
this_prefix = indents + [glyphs.mid]
next_prefix = indents + [glyphs.within_tree]
if node is Ellipsis:
label = " ..."
suffix = ""
children = []
else:
if label_attr is not None:
label = str(graph.nodes[node].get(label_attr, node))
else:
label = str(node)
# Determine if we want to show the children of this node.
if collapse_attr is not None:
collapse = graph.nodes[node].get(collapse_attr, False)
else:
collapse = False
# Determine:
# (1) children to traverse into after showing this node.
# (2) parents to immediately show to the right of this node.
if is_directed:
# In the directed case we must show every successor node
# note: it may be skipped later, but we don't have that
# information here.
children = list(succ[node])
# In the directed case we must show every predecessor
# except for parent we directly traversed from.
handled_parents = {parent}
else:
# Showing only the unseen children results in a more
# concise representation for the undirected case.
children = [
child for child in succ[node] if child not in seen_nodes
]
# In the undirected case, parents are also children, so we
# only need to immediately show the ones we can no longer
# traverse
handled_parents = {*children, parent}
if max_depth is not None and len(indents) == max_depth - 1:
# Use ellipsis to indicate we have reached maximum depth
if children:
children = [Ellipsis]
handled_parents = {parent}
if collapse:
# Collapsing a node is the same as reaching maximum depth
if children:
children = [Ellipsis]
handled_parents = {parent}
# The other parents are other predecessors of this node that
# are not handled elsewhere.
other_parents = [p for p in pred[node] if p not in handled_parents]
if other_parents:
if label_attr is not None:
other_parents_labels = ", ".join(
[
str(graph.nodes[p].get(label_attr, p))
for p in other_parents
]
)
else:
other_parents_labels = ", ".join(
[str(p) for p in other_parents]
)
suffix = " ".join(["", glyphs.backedge, other_parents_labels])
else:
suffix = ""
# Emit the line for this node, this will be called for each node
# exactly once.
if this_vertical:
yield "".join(this_prefix + [glyphs.vertical_edge])
yield "".join(this_prefix + [label, suffix])
if vertical_chains:
if is_directed:
num_children = len(set(children))
else:
num_children = len(set(children) - {parent})
# The next node can be drawn vertically if it is the only
# remaining child of this node.
next_is_vertical = num_children == 1
else:
next_is_vertical = False
# Push children on the stack in reverse order so they are popped in
# the original order.
for idx, child in enumerate(children[::-1]):
next_islast = idx == 0
try_frame = StackFrame(
node, child, next_prefix, next_islast, next_is_vertical
)
stack.append(try_frame)
[docs]
@open_file(1, "w")
def write_network_text(
graph,
path=None,
with_labels=True,
sources=None,
max_depth=None,
ascii_only=False,
end="\n",
vertical_chains=False,
):
"""Creates a nice text representation of a graph
This works via a depth-first traversal of the graph and writing a line for
each unique node encountered. Non-tree edges are written to the right of
each node, and connection to a non-tree edge is indicated with an ellipsis.
This representation works best when the input graph is a forest, but any
graph can be represented.
Parameters
----------
graph : nx.DiGraph | nx.Graph
Graph to represent
path : string or file or callable or None
Filename or file handle for data output.
if a function, then it will be called for each generated line.
if None, this will default to "sys.stdout.write"
with_labels : bool | str
If True will use the "label" attribute of a node to display if it
exists otherwise it will use the node value itself. If given as a
string, then that attribute name will be used instead of "label".
Defaults to True.
sources : List
Specifies which nodes to start traversal from. Note: nodes that are not
reachable from one of these sources may not be shown. If unspecified,
the minimal set of nodes needed to reach all others will be used.
max_depth : int | None
The maximum depth to traverse before stopping. Defaults to None.
ascii_only : Boolean
If True only ASCII characters are used to construct the visualization
end : string
The line ending character
vertical_chains : Boolean
If True, chains of nodes will be drawn vertically when possible.
Examples
--------
>>> graph = nx.balanced_tree(r=2, h=2, create_using=nx.DiGraph)
>>> nx.write_network_text(graph)
╙── 0
├─╼ 1
│ ├─╼ 3
│ └─╼ 4
└─╼ 2
├─╼ 5
└─╼ 6
>>> # A near tree with one non-tree edge
>>> graph.add_edge(5, 1)
>>> nx.write_network_text(graph)
╙── 0
├─╼ 1 ╾ 5
│ ├─╼ 3
│ └─╼ 4
└─╼ 2
├─╼ 5
│ └─╼ ...
└─╼ 6
>>> graph = nx.cycle_graph(5)
>>> nx.write_network_text(graph)
╙── 0
├── 1
│ └── 2
│ └── 3
│ └── 4 ─ 0
└── ...
>>> graph = nx.cycle_graph(5, nx.DiGraph)
>>> nx.write_network_text(graph, vertical_chains=True)
╙── 0 ╾ 4
╽
1
╽
2
╽
3
╽
4
└─╼ ...
>>> nx.write_network_text(graph, vertical_chains=True, ascii_only=True)
+-- 0 <- 4
!
1
!
2
!
3
!
4
L-> ...
>>> graph = nx.generators.barbell_graph(4, 2)
>>> nx.write_network_text(graph, vertical_chains=False)
╙── 4
├── 5
│ └── 6
│ ├── 7
│ │ ├── 8 ─ 6
│ │ │ └── 9 ─ 6, 7
│ │ └── ...
│ └── ...
└── 3
├── 0
│ ├── 1 ─ 3
│ │ └── 2 ─ 0, 3
│ └── ...
└── ...
>>> nx.write_network_text(graph, vertical_chains=True)
╙── 4
├── 5
│ │
│ 6
│ ├── 7
│ │ ├── 8 ─ 6
│ │ │ │
│ │ │ 9 ─ 6, 7
│ │ └── ...
│ └── ...
└── 3
├── 0
│ ├── 1 ─ 3
│ │ │
│ │ 2 ─ 0, 3
│ └── ...
└── ...
>>> graph = nx.complete_graph(5, create_using=nx.Graph)
>>> nx.write_network_text(graph)
╙── 0
├── 1
│ ├── 2 ─ 0
│ │ ├── 3 ─ 0, 1
│ │ │ └── 4 ─ 0, 1, 2
│ │ └── ...
│ └── ...
└── ...
>>> graph = nx.complete_graph(3, create_using=nx.DiGraph)
>>> nx.write_network_text(graph)
╙── 0 ╾ 1, 2
├─╼ 1 ╾ 2
│ ├─╼ 2 ╾ 0
│ │ └─╼ ...
│ └─╼ ...
└─╼ ...
"""
if path is None:
# The path is unspecified, write to stdout
_write = sys.stdout.write
elif hasattr(path, "write"):
# The path is already an open file
_write = path.write
elif callable(path):
# The path is a custom callable
_write = path
else:
raise TypeError(type(path))
for line in generate_network_text(
graph,
with_labels=with_labels,
sources=sources,
max_depth=max_depth,
ascii_only=ascii_only,
vertical_chains=vertical_chains,
):
_write(line + end)
def _find_sources(graph):
"""
Determine a minimal set of nodes such that the entire graph is reachable
"""
# For each connected part of the graph, choose at least
# one node as a starting point, preferably without a parent
if graph.is_directed():
# Choose one node from each SCC with minimum in_degree
sccs = list(nx.strongly_connected_components(graph))
# condensing the SCCs forms a dag, the nodes in this graph with
# 0 in-degree correspond to the SCCs from which the minimum set
# of nodes from which all other nodes can be reached.
scc_graph = nx.condensation(graph, sccs)
supernode_to_nodes = {sn: [] for sn in scc_graph.nodes()}
# Note: the order of mapping differs between pypy and cpython
# so we have to loop over graph nodes for consistency
mapping = scc_graph.graph["mapping"]
for n in graph.nodes:
sn = mapping[n]
supernode_to_nodes[sn].append(n)
sources = []
for sn in scc_graph.nodes():
if scc_graph.in_degree[sn] == 0:
scc = supernode_to_nodes[sn]
node = min(scc, key=lambda n: graph.in_degree[n])
sources.append(node)
else:
# For undirected graph, the entire graph will be reachable as
# long as we consider one node from every connected component
sources = [
min(cc, key=lambda n: graph.degree[n])
for cc in nx.connected_components(graph)
]
sources = sorted(sources, key=lambda n: graph.degree[n])
return sources
def _parse_network_text(lines):
"""Reconstructs a graph from a network text representation.
This is mainly used for testing. Network text is for display, not
serialization, as such this cannot parse all network text representations
because node labels can be ambiguous with the glyphs and indentation used
to represent edge structure. Additionally, there is no way to determine if
disconnected graphs were originally directed or undirected.
Parameters
----------
lines : list or iterator of strings
Input data in network text format
Returns
-------
G: NetworkX graph
The graph corresponding to the lines in network text format.
"""
from itertools import chain
from typing import Any, NamedTuple, Union
class ParseStackFrame(NamedTuple):
node: Any
indent: int
has_vertical_child: int | None
initial_line_iter = iter(lines)
is_ascii = None
is_directed = None
##############
# Initial Pass
##############
# Do an initial pass over the lines to determine what type of graph it is.
# Remember what these lines were, so we can reiterate over them in the
# parsing pass.
initial_lines = []
try:
first_line = next(initial_line_iter)
except StopIteration:
...
else:
initial_lines.append(first_line)
# The first character indicates if it is an ASCII or UTF graph
first_char = first_line[0]
if first_char in {
UtfBaseGlyphs.empty,
UtfBaseGlyphs.newtree_mid[0],
UtfBaseGlyphs.newtree_last[0],
}:
is_ascii = False
elif first_char in {
AsciiBaseGlyphs.empty,
AsciiBaseGlyphs.newtree_mid[0],
AsciiBaseGlyphs.newtree_last[0],
}:
is_ascii = True
else:
raise AssertionError(f"Unexpected first character: {first_char}")
if is_ascii:
directed_glyphs = AsciiDirectedGlyphs.as_dict()
undirected_glyphs = AsciiUndirectedGlyphs.as_dict()
else:
directed_glyphs = UtfDirectedGlyphs.as_dict()
undirected_glyphs = UtfUndirectedGlyphs.as_dict()
# For both directed / undirected glyphs, determine which glyphs never
# appear as substrings in the other undirected / directed glyphs. Glyphs
# with this property unambiguously indicates if a graph is directed /
# undirected.
directed_items = set(directed_glyphs.values())
undirected_items = set(undirected_glyphs.values())
unambiguous_directed_items = []
for item in directed_items:
other_items = undirected_items
other_supersets = [other for other in other_items if item in other]
if not other_supersets:
unambiguous_directed_items.append(item)
unambiguous_undirected_items = []
for item in undirected_items:
other_items = directed_items
other_supersets = [other for other in other_items if item in other]
if not other_supersets:
unambiguous_undirected_items.append(item)
for line in initial_line_iter:
initial_lines.append(line)
if any(item in line for item in unambiguous_undirected_items):
is_directed = False
break
elif any(item in line for item in unambiguous_directed_items):
is_directed = True
break
if is_directed is None:
# Not enough information to determine, choose undirected by default
is_directed = False
glyphs = directed_glyphs if is_directed else undirected_glyphs
# the backedge symbol by itself can be ambiguous, but with spaces around it
# becomes unambiguous.
backedge_symbol = " " + glyphs["backedge"] + " "
# Reconstruct an iterator over all of the lines.
parsing_line_iter = chain(initial_lines, initial_line_iter)
##############
# Parsing Pass
##############
edges = []
nodes = []
is_empty = None
noparent = object() # sentinel value
# keep a stack of previous nodes that could be parents of subsequent nodes
stack = [ParseStackFrame(noparent, -1, None)]
for line in parsing_line_iter:
if line == glyphs["empty"]:
# If the line is the empty glyph, we are done.
# There shouldn't be anything else after this.
is_empty = True
continue
if backedge_symbol in line:
# This line has one or more backedges, separate those out
node_part, backedge_part = line.split(backedge_symbol)
backedge_nodes = [u.strip() for u in backedge_part.split(", ")]
# Now the node can be parsed
node_part = node_part.rstrip()
prefix, node = node_part.rsplit(" ", 1)
node = node.strip()
# Add the backedges to the edge list
edges.extend([(u, node) for u in backedge_nodes])
else:
# No backedge, the tail of this line is the node
prefix, node = line.rsplit(" ", 1)
node = node.strip()
prev = stack.pop()
if node in glyphs["vertical_edge"]:
# Previous node is still the previous node, but we know it will
# have exactly one child, which will need to have its nesting level
# adjusted.
modified_prev = ParseStackFrame(
prev.node,
prev.indent,
True,
)
stack.append(modified_prev)
continue
# The length of the string before the node characters give us a hint
# about our nesting level. The only case where this doesn't work is
# when there are vertical chains, which is handled explicitly.
indent = len(prefix)
curr = ParseStackFrame(node, indent, None)
if prev.has_vertical_child:
# In this case we know prev must be the parent of our current line,
# so we don't have to search the stack. (which is good because the
# indentation check wouldn't work in this case).
...
else:
# If the previous node nesting-level is greater than the current
# nodes nesting-level than the previous node was the end of a path,
# and is not our parent. We can safely pop nodes off the stack
# until we find one with a comparable nesting-level, which is our
# parent.
while curr.indent <= prev.indent:
prev = stack.pop()
if node == "...":
# The current previous node is no longer a valid parent,
# keep it popped from the stack.
stack.append(prev)
else:
# The previous and current nodes may still be parents, so add them
# back onto the stack.
stack.append(prev)
stack.append(curr)
# Add the node and the edge to its parent to the node / edge lists.
nodes.append(curr.node)
if prev.node is not noparent:
edges.append((prev.node, curr.node))
if is_empty:
# Sanity check
assert len(nodes) == 0
# Reconstruct the graph
cls = nx.DiGraph if is_directed else nx.Graph
new = cls()
new.add_nodes_from(nodes)
new.add_edges_from(edges)
return new
