An independent set on a graph is a set of nodes where no two nodes in the set are neighbors. A maximal independent set (MIS) is an independent set where no node can be added without having two neighbors in the set.
This problem is interesting when we are considering a graph that can be dynamically updated. Possible updates are removal/insertion of an edge/vertex.
Instead of recomputing the MIS from scratch, it is desirable to re-use information from the old solution. Ideally, this leads to faster solutions after an update.
This project contains several different algorithms that compute a MIS on a dynamically changeable graph.
-
TrivialMIS
The intuitive approach to building a MIS. It completely recomputes the MIS after every update
-
SimpleMIS
Each vertex keeps a count of how many neighbors are in the MIS. Nodes are added/removed from the MIS based on this counter.
-
ImprovedIncrementalMIS
A small change to SimpleMIS, that only implements edge insertions. In the case that both nodes of the edge are in the MIS, the node with lower degree will be removed.
-
ImprovedDynamicMIS
A fully dynamic algorithm, that classifies nodes as either heavy or light based on their degree and performs different operations accordingly to achieve lower amortized costs.
-
ImplicitMIS
In a relaxed model we need not maintain an explicit version of the MIS. Instead this algorithms only saves an independent set. If a node not in this set is part of the MIS is decided lazily.
The code was tested using Python 3.8.2.
The requirements are listed in requirements.txt.
The code uses the networkx library as the backend for graph operations.
Furthermore numpy.random is used to generate reproducible random numbers.
The code can be imported as follows:
import dynamic_mis as dm
The implemented algorithms all share a common interface. To create a new instance of an algorithm on a graph call
graph = *your networkx graph*
algo = dm.TrivialMIS(graph)
where graph is a networkx graph.
Information about the maximal independet set is exposed via two member functions:
algo.is_in_mis(node) # Boolean
mis = algo.get_mis() # set object
To perform updates to the graph one of these four functions can be used:
algo.insert_edge(u, v)
algo.remove_edge(u, v)
algo.insert_node(v, edges) # The edges parameter is optional
algo.remove_node(v)Note that the complexity of these function calls depends on the specific algorithm.
The code has been benchmarked using different networks from the Koblenz Network Collection.
To execute a specific benchmark uncomment the relevant section in benchmark.py.
Then execute
python3 -m dynamic_mis.benchmark data_dir/
data_dir is the path to the directory where the network files are located.
The edge files should contain only edge data and no additional lines with metadata.