425 lines
15 KiB
Python
425 lines
15 KiB
Python
"""
|
|
Highest-label preflow-push algorithm for maximum flow problems.
|
|
"""
|
|
|
|
from collections import deque
|
|
from itertools import islice
|
|
import networkx as nx
|
|
from ...utils import arbitrary_element
|
|
from .utils import build_residual_network
|
|
from .utils import CurrentEdge
|
|
from .utils import detect_unboundedness
|
|
from .utils import GlobalRelabelThreshold
|
|
from .utils import Level
|
|
|
|
__all__ = ["preflow_push"]
|
|
|
|
|
|
def preflow_push_impl(G, s, t, capacity, residual, global_relabel_freq, value_only):
|
|
"""Implementation of the highest-label preflow-push algorithm.
|
|
"""
|
|
if s not in G:
|
|
raise nx.NetworkXError(f"node {str(s)} not in graph")
|
|
if t not in G:
|
|
raise nx.NetworkXError(f"node {str(t)} not in graph")
|
|
if s == t:
|
|
raise nx.NetworkXError("source and sink are the same node")
|
|
|
|
if global_relabel_freq is None:
|
|
global_relabel_freq = 0
|
|
if global_relabel_freq < 0:
|
|
raise nx.NetworkXError("global_relabel_freq must be nonnegative.")
|
|
|
|
if residual is None:
|
|
R = build_residual_network(G, capacity)
|
|
else:
|
|
R = residual
|
|
|
|
detect_unboundedness(R, s, t)
|
|
|
|
R_nodes = R.nodes
|
|
R_pred = R.pred
|
|
R_succ = R.succ
|
|
|
|
# Initialize/reset the residual network.
|
|
for u in R:
|
|
R_nodes[u]["excess"] = 0
|
|
for e in R_succ[u].values():
|
|
e["flow"] = 0
|
|
|
|
def reverse_bfs(src):
|
|
"""Perform a reverse breadth-first search from src in the residual
|
|
network.
|
|
"""
|
|
heights = {src: 0}
|
|
q = deque([(src, 0)])
|
|
while q:
|
|
u, height = q.popleft()
|
|
height += 1
|
|
for v, attr in R_pred[u].items():
|
|
if v not in heights and attr["flow"] < attr["capacity"]:
|
|
heights[v] = height
|
|
q.append((v, height))
|
|
return heights
|
|
|
|
# Initialize heights of the nodes.
|
|
heights = reverse_bfs(t)
|
|
|
|
if s not in heights:
|
|
# t is not reachable from s in the residual network. The maximum flow
|
|
# must be zero.
|
|
R.graph["flow_value"] = 0
|
|
return R
|
|
|
|
n = len(R)
|
|
# max_height represents the height of the highest level below level n with
|
|
# at least one active node.
|
|
max_height = max(heights[u] for u in heights if u != s)
|
|
heights[s] = n
|
|
|
|
grt = GlobalRelabelThreshold(n, R.size(), global_relabel_freq)
|
|
|
|
# Initialize heights and 'current edge' data structures of the nodes.
|
|
for u in R:
|
|
R_nodes[u]["height"] = heights[u] if u in heights else n + 1
|
|
R_nodes[u]["curr_edge"] = CurrentEdge(R_succ[u])
|
|
|
|
def push(u, v, flow):
|
|
"""Push flow units of flow from u to v.
|
|
"""
|
|
R_succ[u][v]["flow"] += flow
|
|
R_succ[v][u]["flow"] -= flow
|
|
R_nodes[u]["excess"] -= flow
|
|
R_nodes[v]["excess"] += flow
|
|
|
|
# The maximum flow must be nonzero now. Initialize the preflow by
|
|
# saturating all edges emanating from s.
|
|
for u, attr in R_succ[s].items():
|
|
flow = attr["capacity"]
|
|
if flow > 0:
|
|
push(s, u, flow)
|
|
|
|
# Partition nodes into levels.
|
|
levels = [Level() for i in range(2 * n)]
|
|
for u in R:
|
|
if u != s and u != t:
|
|
level = levels[R_nodes[u]["height"]]
|
|
if R_nodes[u]["excess"] > 0:
|
|
level.active.add(u)
|
|
else:
|
|
level.inactive.add(u)
|
|
|
|
def activate(v):
|
|
"""Move a node from the inactive set to the active set of its level.
|
|
"""
|
|
if v != s and v != t:
|
|
level = levels[R_nodes[v]["height"]]
|
|
if v in level.inactive:
|
|
level.inactive.remove(v)
|
|
level.active.add(v)
|
|
|
|
def relabel(u):
|
|
"""Relabel a node to create an admissible edge.
|
|
"""
|
|
grt.add_work(len(R_succ[u]))
|
|
return (
|
|
min(
|
|
R_nodes[v]["height"]
|
|
for v, attr in R_succ[u].items()
|
|
if attr["flow"] < attr["capacity"]
|
|
)
|
|
+ 1
|
|
)
|
|
|
|
def discharge(u, is_phase1):
|
|
"""Discharge a node until it becomes inactive or, during phase 1 (see
|
|
below), its height reaches at least n. The node is known to have the
|
|
largest height among active nodes.
|
|
"""
|
|
height = R_nodes[u]["height"]
|
|
curr_edge = R_nodes[u]["curr_edge"]
|
|
# next_height represents the next height to examine after discharging
|
|
# the current node. During phase 1, it is capped to below n.
|
|
next_height = height
|
|
levels[height].active.remove(u)
|
|
while True:
|
|
v, attr = curr_edge.get()
|
|
if height == R_nodes[v]["height"] + 1 and attr["flow"] < attr["capacity"]:
|
|
flow = min(R_nodes[u]["excess"], attr["capacity"] - attr["flow"])
|
|
push(u, v, flow)
|
|
activate(v)
|
|
if R_nodes[u]["excess"] == 0:
|
|
# The node has become inactive.
|
|
levels[height].inactive.add(u)
|
|
break
|
|
try:
|
|
curr_edge.move_to_next()
|
|
except StopIteration:
|
|
# We have run off the end of the adjacency list, and there can
|
|
# be no more admissible edges. Relabel the node to create one.
|
|
height = relabel(u)
|
|
if is_phase1 and height >= n - 1:
|
|
# Although the node is still active, with a height at least
|
|
# n - 1, it is now known to be on the s side of the minimum
|
|
# s-t cut. Stop processing it until phase 2.
|
|
levels[height].active.add(u)
|
|
break
|
|
# The first relabel operation after global relabeling may not
|
|
# increase the height of the node since the 'current edge' data
|
|
# structure is not rewound. Use height instead of (height - 1)
|
|
# in case other active nodes at the same level are missed.
|
|
next_height = height
|
|
R_nodes[u]["height"] = height
|
|
return next_height
|
|
|
|
def gap_heuristic(height):
|
|
"""Apply the gap heuristic.
|
|
"""
|
|
# Move all nodes at levels (height + 1) to max_height to level n + 1.
|
|
for level in islice(levels, height + 1, max_height + 1):
|
|
for u in level.active:
|
|
R_nodes[u]["height"] = n + 1
|
|
for u in level.inactive:
|
|
R_nodes[u]["height"] = n + 1
|
|
levels[n + 1].active.update(level.active)
|
|
level.active.clear()
|
|
levels[n + 1].inactive.update(level.inactive)
|
|
level.inactive.clear()
|
|
|
|
def global_relabel(from_sink):
|
|
"""Apply the global relabeling heuristic.
|
|
"""
|
|
src = t if from_sink else s
|
|
heights = reverse_bfs(src)
|
|
if not from_sink:
|
|
# s must be reachable from t. Remove t explicitly.
|
|
del heights[t]
|
|
max_height = max(heights.values())
|
|
if from_sink:
|
|
# Also mark nodes from which t is unreachable for relabeling. This
|
|
# serves the same purpose as the gap heuristic.
|
|
for u in R:
|
|
if u not in heights and R_nodes[u]["height"] < n:
|
|
heights[u] = n + 1
|
|
else:
|
|
# Shift the computed heights because the height of s is n.
|
|
for u in heights:
|
|
heights[u] += n
|
|
max_height += n
|
|
del heights[src]
|
|
for u, new_height in heights.items():
|
|
old_height = R_nodes[u]["height"]
|
|
if new_height != old_height:
|
|
if u in levels[old_height].active:
|
|
levels[old_height].active.remove(u)
|
|
levels[new_height].active.add(u)
|
|
else:
|
|
levels[old_height].inactive.remove(u)
|
|
levels[new_height].inactive.add(u)
|
|
R_nodes[u]["height"] = new_height
|
|
return max_height
|
|
|
|
# Phase 1: Find the maximum preflow by pushing as much flow as possible to
|
|
# t.
|
|
|
|
height = max_height
|
|
while height > 0:
|
|
# Discharge active nodes in the current level.
|
|
while True:
|
|
level = levels[height]
|
|
if not level.active:
|
|
# All active nodes in the current level have been discharged.
|
|
# Move to the next lower level.
|
|
height -= 1
|
|
break
|
|
# Record the old height and level for the gap heuristic.
|
|
old_height = height
|
|
old_level = level
|
|
u = arbitrary_element(level.active)
|
|
height = discharge(u, True)
|
|
if grt.is_reached():
|
|
# Global relabeling heuristic: Recompute the exact heights of
|
|
# all nodes.
|
|
height = global_relabel(True)
|
|
max_height = height
|
|
grt.clear_work()
|
|
elif not old_level.active and not old_level.inactive:
|
|
# Gap heuristic: If the level at old_height is empty (a 'gap'),
|
|
# a minimum cut has been identified. All nodes with heights
|
|
# above old_height can have their heights set to n + 1 and not
|
|
# be further processed before a maximum preflow is found.
|
|
gap_heuristic(old_height)
|
|
height = old_height - 1
|
|
max_height = height
|
|
else:
|
|
# Update the height of the highest level with at least one
|
|
# active node.
|
|
max_height = max(max_height, height)
|
|
|
|
# A maximum preflow has been found. The excess at t is the maximum flow
|
|
# value.
|
|
if value_only:
|
|
R.graph["flow_value"] = R_nodes[t]["excess"]
|
|
return R
|
|
|
|
# Phase 2: Convert the maximum preflow into a maximum flow by returning the
|
|
# excess to s.
|
|
|
|
# Relabel all nodes so that they have accurate heights.
|
|
height = global_relabel(False)
|
|
grt.clear_work()
|
|
|
|
# Continue to discharge the active nodes.
|
|
while height > n:
|
|
# Discharge active nodes in the current level.
|
|
while True:
|
|
level = levels[height]
|
|
if not level.active:
|
|
# All active nodes in the current level have been discharged.
|
|
# Move to the next lower level.
|
|
height -= 1
|
|
break
|
|
u = arbitrary_element(level.active)
|
|
height = discharge(u, False)
|
|
if grt.is_reached():
|
|
# Global relabeling heuristic.
|
|
height = global_relabel(False)
|
|
grt.clear_work()
|
|
|
|
R.graph["flow_value"] = R_nodes[t]["excess"]
|
|
return R
|
|
|
|
|
|
def preflow_push(
|
|
G, s, t, capacity="capacity", residual=None, global_relabel_freq=1, value_only=False
|
|
):
|
|
r"""Find a maximum single-commodity flow using the highest-label
|
|
preflow-push algorithm.
|
|
|
|
This function returns the residual network resulting after computing
|
|
the maximum flow. See below for details about the conventions
|
|
NetworkX uses for defining residual networks.
|
|
|
|
This algorithm has a running time of $O(n^2 \sqrt{m})$ for $n$ nodes and
|
|
$m$ edges.
|
|
|
|
|
|
Parameters
|
|
----------
|
|
G : NetworkX graph
|
|
Edges of the graph are expected to have an attribute called
|
|
'capacity'. If this attribute is not present, the edge is
|
|
considered to have infinite capacity.
|
|
|
|
s : node
|
|
Source node for the flow.
|
|
|
|
t : node
|
|
Sink node for the flow.
|
|
|
|
capacity : string
|
|
Edges of the graph G are expected to have an attribute capacity
|
|
that indicates how much flow the edge can support. If this
|
|
attribute is not present, the edge is considered to have
|
|
infinite capacity. Default value: 'capacity'.
|
|
|
|
residual : NetworkX graph
|
|
Residual network on which the algorithm is to be executed. If None, a
|
|
new residual network is created. Default value: None.
|
|
|
|
global_relabel_freq : integer, float
|
|
Relative frequency of applying the global relabeling heuristic to speed
|
|
up the algorithm. If it is None, the heuristic is disabled. Default
|
|
value: 1.
|
|
|
|
value_only : bool
|
|
If False, compute a maximum flow; otherwise, compute a maximum preflow
|
|
which is enough for computing the maximum flow value. Default value:
|
|
False.
|
|
|
|
Returns
|
|
-------
|
|
R : NetworkX DiGraph
|
|
Residual network after computing the maximum flow.
|
|
|
|
Raises
|
|
------
|
|
NetworkXError
|
|
The algorithm does not support MultiGraph and MultiDiGraph. If
|
|
the input graph is an instance of one of these two classes, a
|
|
NetworkXError is raised.
|
|
|
|
NetworkXUnbounded
|
|
If the graph has a path of infinite capacity, the value of a
|
|
feasible flow on the graph is unbounded above and the function
|
|
raises a NetworkXUnbounded.
|
|
|
|
See also
|
|
--------
|
|
:meth:`maximum_flow`
|
|
:meth:`minimum_cut`
|
|
:meth:`edmonds_karp`
|
|
:meth:`shortest_augmenting_path`
|
|
|
|
Notes
|
|
-----
|
|
The residual network :samp:`R` from an input graph :samp:`G` has the
|
|
same nodes as :samp:`G`. :samp:`R` is a DiGraph that contains a pair
|
|
of edges :samp:`(u, v)` and :samp:`(v, u)` iff :samp:`(u, v)` is not a
|
|
self-loop, and at least one of :samp:`(u, v)` and :samp:`(v, u)` exists
|
|
in :samp:`G`. For each node :samp:`u` in :samp:`R`,
|
|
:samp:`R.nodes[u]['excess']` represents the difference between flow into
|
|
:samp:`u` and flow out of :samp:`u`.
|
|
|
|
For each edge :samp:`(u, v)` in :samp:`R`, :samp:`R[u][v]['capacity']`
|
|
is equal to the capacity of :samp:`(u, v)` in :samp:`G` if it exists
|
|
in :samp:`G` or zero otherwise. If the capacity is infinite,
|
|
:samp:`R[u][v]['capacity']` will have a high arbitrary finite value
|
|
that does not affect the solution of the problem. This value is stored in
|
|
:samp:`R.graph['inf']`. For each edge :samp:`(u, v)` in :samp:`R`,
|
|
:samp:`R[u][v]['flow']` represents the flow function of :samp:`(u, v)` and
|
|
satisfies :samp:`R[u][v]['flow'] == -R[v][u]['flow']`.
|
|
|
|
The flow value, defined as the total flow into :samp:`t`, the sink, is
|
|
stored in :samp:`R.graph['flow_value']`. Reachability to :samp:`t` using
|
|
only edges :samp:`(u, v)` such that
|
|
:samp:`R[u][v]['flow'] < R[u][v]['capacity']` induces a minimum
|
|
:samp:`s`-:samp:`t` cut.
|
|
|
|
Examples
|
|
--------
|
|
>>> from networkx.algorithms.flow import preflow_push
|
|
|
|
The functions that implement flow algorithms and output a residual
|
|
network, such as this one, are not imported to the base NetworkX
|
|
namespace, so you have to explicitly import them from the flow package.
|
|
|
|
>>> G = nx.DiGraph()
|
|
>>> G.add_edge("x", "a", capacity=3.0)
|
|
>>> G.add_edge("x", "b", capacity=1.0)
|
|
>>> G.add_edge("a", "c", capacity=3.0)
|
|
>>> G.add_edge("b", "c", capacity=5.0)
|
|
>>> G.add_edge("b", "d", capacity=4.0)
|
|
>>> G.add_edge("d", "e", capacity=2.0)
|
|
>>> G.add_edge("c", "y", capacity=2.0)
|
|
>>> G.add_edge("e", "y", capacity=3.0)
|
|
>>> R = preflow_push(G, "x", "y")
|
|
>>> flow_value = nx.maximum_flow_value(G, "x", "y")
|
|
>>> flow_value == R.graph["flow_value"]
|
|
True
|
|
>>> # preflow_push also stores the maximum flow value
|
|
>>> # in the excess attribute of the sink node t
|
|
>>> flow_value == R.nodes["y"]["excess"]
|
|
True
|
|
>>> # For some problems, you might only want to compute a
|
|
>>> # maximum preflow.
|
|
>>> R = preflow_push(G, "x", "y", value_only=True)
|
|
>>> flow_value == R.graph["flow_value"]
|
|
True
|
|
>>> flow_value == R.nodes["y"]["excess"]
|
|
True
|
|
|
|
"""
|
|
R = preflow_push_impl(G, s, t, capacity, residual, global_relabel_freq, value_only)
|
|
R.graph["algorithm"] = "preflow_push"
|
|
return R
|