Strongly connected components (#2114)

* Implement strongly connected components for graph algorithms

* fixup! Format Python code with psf/black push

* Delete trailing whitespace

* updating DIRECTORY.md

* Add doctests and typehints

* Remove unnecessary comments, change variable names

* fixup! Format Python code with psf/black push

* Change undefined variable's name

* Apply suggestions from code review

Co-authored-by: Christian Clauss <cclauss@me.com>

Co-authored-by: github-actions <${GITHUB_ACTOR}@users.noreply.github.com>
Co-authored-by: Christian Clauss <cclauss@me.com>
This commit is contained in:
Nika Losaberidze 2020-06-17 20:16:54 +04:00 committed by GitHub
parent 2bbdc3bfe7
commit fb3a228d26
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 106 additions and 0 deletions

View File

@ -261,6 +261,7 @@
* [Page Rank](https://github.com/TheAlgorithms/Python/blob/master/graphs/page_rank.py) * [Page Rank](https://github.com/TheAlgorithms/Python/blob/master/graphs/page_rank.py)
* [Prim](https://github.com/TheAlgorithms/Python/blob/master/graphs/prim.py) * [Prim](https://github.com/TheAlgorithms/Python/blob/master/graphs/prim.py)
* [Scc Kosaraju](https://github.com/TheAlgorithms/Python/blob/master/graphs/scc_kosaraju.py) * [Scc Kosaraju](https://github.com/TheAlgorithms/Python/blob/master/graphs/scc_kosaraju.py)
* [Strongly Connected Components](https://github.com/TheAlgorithms/Python/blob/master/graphs/strongly_connected_components.py)
* [Tarjans Scc](https://github.com/TheAlgorithms/Python/blob/master/graphs/tarjans_scc.py) * [Tarjans Scc](https://github.com/TheAlgorithms/Python/blob/master/graphs/tarjans_scc.py)
## Greedy Method ## Greedy Method

View File

@ -0,0 +1,105 @@
"""
https://en.wikipedia.org/wiki/Strongly_connected_component
Finding strongly connected components in directed graph
"""
test_graph_1 = {
0: [2, 3],
1: [0],
2: [1],
3: [4],
4: [],
}
test_graph_2 = {
0: [1, 2, 3],
1: [2],
2: [0],
3: [4],
4: [5],
5: [3],
}
def topology_sort(graph: dict, vert: int, visited: list) -> list:
"""
Use depth first search to sort graph
At this time graph is the same as input
>>> topology_sort(test_graph_1, 0, 5 * [False])
[1, 2, 4, 3, 0]
>>> topology_sort(test_graph_2, 0, 6 * [False])
[2, 1, 5, 4, 3, 0]
"""
visited[vert] = True
order = []
for neighbour in graph[vert]:
if not visited[neighbour]:
order += topology_sort(graph, neighbour, visited)
order.append(vert)
return order
def find_components(reversed_graph: dict, vert: int, visited: list) -> list:
"""
Use depth first search to find strongliy connected
vertices. Now graph is reversed
>>> find_components({0: [1], 1: [2], 2: [0]}, 0, 5 * [False])
[0, 1, 2]
>>> find_components({0: [2], 1: [0], 2: [0, 1]}, 0, 6 * [False])
[0, 2, 1]
"""
visited[vert] = True
component = [vert]
for neighbour in reversed_graph[vert]:
if not visited[neighbour]:
component += find_components(reversed_graph, neighbour, visited)
return component
def strongly_connected_components(graph: dict) -> list:
"""
This function takes graph as a parameter
and then returns the list of strongly connected components
>>> strongly_connected_components(test_graph_1)
[[0, 1, 2], [3], [4]]
>>> strongly_connected_components(test_graph_2)
[[0, 2, 1], [3, 5, 4]]
"""
visited = len(graph) * [False]
reversed_graph = {vert: [] for vert in range(len(graph))}
for vert, neighbours in graph.items():
for neighbour in neighbours:
reversed_graph[neighbour].append(vert)
order = []
for i, was_visited in enumerate(visited):
if not was_visited:
order += topology_sort(graph, i, visited)
components_list = []
visited = len(graph) * [False]
for i in range(len(graph)):
vert = order[len(graph) - i - 1]
if not visited[vert]:
component = find_components(reversed_graph, vert, visited)
components_list.append(component)
return components_list
if __name__ == "__main__":
import doctest
doctest.testmod()