Louvain documentation

Installation

In short: pip install louvain. Alternatively, use Anaconda and get the conda packages from the conda-forge channel, which supports both Unix, Mac OS and Windows.

For Unix like systems it is possible to install from source. For Windows this is overly complicated, and you are recommended to use the binary wheels. There are two things that are needed by this package: the igraph C core library and the python-igraph python package. For both, please see http://igraph.org.

Make sure you have all necessary tools for compilation. In Ubuntu this can be installed using sudo apt-get install build-essential, please refer to the documentation for your specific system. Make sure that not only gcc is installed, but also g++, as the louvain-igraph package is programmed in C++.

You can check if all went well by running a variety of tests using python setup.py test.

There are basically two installation modes, similar to the python-igraph package itself (from which most of the setup.py comes).

1. No C core library is installed yet. The packages will be compiled and linked statically to an automatically downloaded version of the C core library of igraph.

2. A C core library is already installed. In this case, the package will link dynamically to the already installed version. This is probably also the version that is used by the igraph package, but you may want to double check this.

In case the python-igraph package is already installed before, make sure that both use the same versions.

The cleanest setup it to install and compile the C core library yourself (make sure that the header files are also included, e.g. install also the development package from igraph). Then both the python-igraph package, as well as this package are compiled and (dynamically) linked to the same C core library.

Introduction

This package facilitates community detection of networks and builds on the package igraph, referred to as ig throughout this documentation. Although the options in the package are extensive, most people are presumably simply interested in detecting communities with a robust method that works well. This introduction explains how to do that.

For those without patience (and some prior experience), if you simply want to detect communities given a graph G using modularity, you simply use

>>> partition = louvain.find_partition(G, louvain.ModularityVertexPartition);

That’s it.

Why then should you use this package rather than the Louvain algorithm community_multilevel() built into igraph? If you want to use modularity, and you work with a simple undirected, unweighted graph, then indeed you may use the built-in method. For anything else, the functionality is not built-in and this package is for you.

For those less familiar with igraph, let us work out an example more fully. First, we need to import the relevant packages:

>>> import igraph as ig
>>> import louvain

Let us then look at one of the most famous examples of network science: the Zachary karate club (it even has a prize named after it):

>>> G = ig.Graph.Famous('Zachary')

Now detecting communities with modularity is straightforward, as demonstrated earlier:

>>> partition = louvain.find_partition(G, louvain.ModularityVertexPartition)

You can simply plot the results as follows:

>>> ig.plot(partition) 

In this case, the algorithm actually finds the optimal partition (for small graphs like these you can check this using community_optimal_modularity() in the igraph package), but this is generally not the case (although the algorithm should do well). Although this is the optimal partition, it does not correspond to the split in two factions that was observed for this particular network. We can uncover that split in two using a different method: CPMVertexPartition:

>>> partition = louvain.find_partition(G, louvain.CPMVertexPartition,
...                                    resolution_parameter = 0.05);
>>> ig.plot(partition) 

Note that any additional **kwargs passed to find_partition() is passed on to the constructor of the given partition_type. In this case, we can pass the resolution_parameter, but we could also pass weights or node_sizes.

This is the real benefit of using this package: it provides implementations for six different methods (see Reference), and works also on directed and weighted graphs. Finally, it also allows to work with more complex multiplex graphs (see Multiplex).

Advanced

The basic interface explained in the Introduction should provide you enough to start detecting communities. However, perhaps you want to improve the partitions further or want to do some more advanced analysis. In this section, we will explain this in more detail.

Optimiser

Although the package provides simple access to the function find_partition(), there is actually an underlying Optimiser class that is doing the actual work. We can also explicitly construct an Optimiser object:

>>> optimiser = louvain.Optimiser()

The function find_partition() then does nothing else then calling optimise_partition() on the provided partition.

>>> diff = optimiser.optimise_partition(partition)

But optimise_partition() simply tries to improve any provided partition. We can thus try to repeatedly call optimise_partition() to keep on improving the current partition:

>>> G = ig.Graph.Erdos_Renyi(100, p=5./100)
>>> partition = louvain.ModularityVertexPartition(G)
>>> improv = 1
>>> while improv > 0:
...   improv = optimiser.optimise_partition(partition)

Even if a call to optimise_partition() did not improve the current partition, it is still possible that a next call will improve the partition. Of course, if the current partition is already optimal, this will never happen, but it is not possible to decide whether a partition is optimal.

The optimise_partition() itself is built on a basic algorithm: move_nodes(). You can also call this function yourself. For example:

>>> diff = optimiser.move_nodes(partition)

The usual strategy in the Louvain algorithm is then to aggregate the partition and repeat the move_nodes() on the aggregated partition. We can easily repeat that:

>>> partition = louvain.ModularityVertexPartition(G)
>>> while optimiser.move_nodes(partition) > 0:
...   partition = partition.aggregate_partition()

This summarises the whole Louvain algorithm in just three lines of code. Although this finds the final aggregate partition, this leaves it unclear the actual partition on the level of the individual nodes. In order to do that, we need to update the membership based on the aggregate partition, for which we use the function from_coarse_partition().

>>> partition = louvain.ModularityVertexPartition(G)
>>> partition_agg = partition.aggregate_partition()
>>> while optimiser.move_nodes(partition_agg):
...   partition.from_coarse_partition(partition_agg)
...   partition_agg = partition_agg.aggregate_partition()

Now partition_agg contains the aggregate partition and partition contains the actual partition of the original graph G. Of course, partition_agg.quality() == partition.quality() (save some rounding).

The function move_nodes() in turn relies on two key functions of the partition: diff_move() and move_node(). The first calculates the difference when moving a node, and the latter actually moves the node, and updates all necessary internal administration. The move_nodes() then does something as follows

>>> for v in G.vs:
...   best_comm = max(range(len(partition)),
...                   key=lambda c: partition.diff_move(v.index, c))
...   partition.move_node(v.index, best_comm)

The actual implementation is more complicated, but this gives the general idea.

Resolution profile

Some methods accept so-called resolution parameters, such as CPMVertexPartition or RBConfigurationVertexPartition. Although some method may seem to have some ‘natural’ resolution, in reality this is often quite arbitrary. However, the methods implemented here (which depend in a linear way on resolution parameters) allow for an effective scanning of a full range for the resolution parameter. In particular, these methods somehow can be formulated as \(Q = E - \gamma N\) where \(E\) and \(N\) are some other quantities. In the case for CPMVertexPartition for example, \(E = \sum_c m_c\) is the number of internal edges and \(N = \sum_c \binom{n_c}{2}\) is the sum of the internal possible edges. The essential insight for these formulations 1 is that if there is an optimal partition for both \(\gamma_1\) and \(\gamma_2\) then the partition is also optimal for all \(\gamma_1 \leq \gamma \leq \gamma_2\).

Such a resolution profile can be constructed using the Optimiser object.

>>> G = ig.Graph.Famous('Zachary')
>>> optimiser = louvain.Optimiser()
>>> profile = optimiser.resolution_profile(G, louvain.CPMVertexPartition,
...                                        resolution_range=(0,1))

Plotting the resolution parameter versus the total number of internal edges we thus obtain something as follows:

Now profile contains a list of partitions of the specified type (CPMVertexPartition in this case) for resolution parameters at which there was a change. In particular, profile[i] should be better until profile[i+1], or stated otherwise for any resolution parameter between profile[i].resolution_parameter and profile[i+1].resolution_parameter the partition at position i should be better. Of course, there will be some variations because optimise_partition() will find partitions of varying quality. The change points can then also vary for different runs.

This function repeatedly calls optimise_partition() and can therefore require a lot of time. Especially for resolution parameters right around a change point there may be many possible partitions, thus requiring a lot of runs.

References

1

Traag, V. A., Krings, G., & Van Dooren, P. (2013). Significant scales in community structure. Scientific Reports, 3, 2930. 10.1038/srep02930

Multiplex

The implementation of multiplex community detection builds on ideas in 1. The most basic form simply considers two or more graphs which are defined on the same vertex set, but which have differing edge sets. In this context, each node is identified with a single community, and cannot have different communities for different graphs. We call this layers of graphs in this context. This format is actually more flexible than it looks, but you have to construct the layer graphs in a smart way. Instead of having layers of graphs which are always identified on the same vertex set, you could define slices of graphs which do not necessarily have the same vertex set. Using slices we would like to assign a node to a community for each slice, so that the community for a node can be different for different slices, rather than always being the same for all layers. We can translate slices into layers but it is not an easy transformation to grasp fully. But by doing so, we can again rely on the same machinery we developed for dealing with layers.

Throughout the remained of this section, we assume an optimiser has been created:

>>> optimiser = louvain.Optimiser()

Layer multiplex

If we have two graphs which are identified on exactly the same vertex set, we say we have two layers. For example, suppose graph G_telephone contains the communication between friends over the telephone and that the graph G_email contains the communication between friends via mail. The exact same vertex set then means that G_telephone.vs[i] is identical to the node G_email.vs[i]. For each layer we can separately specify the type of partition that we look for. In principle they could be different for each layer, but for now we will assume the type of partition is the same for all layers. The quality of all partitions combined is simply the sum of the individual qualities for the various partitions, weighted by the layer_weight. If we denote by \(q_k\) the quality of layer \(k\) and the weight by \(w_k\), the overall quality is then

\[q = \sum_k w_k q_k.\]

The optimisation algorithm is no different from the standard algorithm. We simply calculate the overall difference of moving a node to another community as the sum of the individual differences in all partitions. The rest (aggregating and repeating on the aggregate partition) simple proceeds as usual.

The most straightforward way to use this is then to use find_partition_multiplex():

>>> membership, improv = louvain.find_partition_multiplex(
...                        [G_telephone, G_email],
...                        louvain.ModularityVertexPartition);

Note

You may need to carefully reflect how you want to weigh the importance of an individual layer. Since the ModularityVertexPartition is normalised by the number of links, you essentially weigh layers the same, independent of the number of links. This may be undesirable, in which case it may be better to use RBConfigurationVertexPartition, which is unnormalised. Alternatively, you may specify different layer_weights.

Similar to the simpler function find_partition(), it is a simple helper function. The function returns a membership vector, because the membership for all layers is identical. You can also control the partitions and optimisation in more detail. Perhaps it is better to use CPMVertexPartition with different resolution parameter for example for different layers of the graph. For example, using email creates a more connected structure because multiple people can be involved in a single mail, which may require a higher resolution parameter for the email graph.

>>> part_telephone = louvain.CPMVertexPartition(
...                    G_telephone, resolution_parameter=0.01);
>>> part_email = louvain.CPMVertexPartition(
...                    G_email, resolution_parameter=0.3);
>>> diff = optimiser.optimise_partition_multiplex(
...                    [part_telephone, part_email]);

Note that part_telephone and part_email contain exactly the same partition, in the sense that part_telephone.membership == part_email.membership. The underlying graph is of course different, and hence the individual quality will also be different.

Some layers may have a more important role in the partition and this can be indicated by the layer_weight. Using half the weight for the email layer for example would be possible as follows:

>>> diff = optimiser.optimise_partition_multiplex(
...   [part_telephone, part_email],
...   layer_weights=[1,0.5]);

Negative links

The layer weights are especially useful when negative links are present, representing for example conflict or animosity. Most methods (except CPM) only accept positive weights. In order to deal with graphs that do have negative links, a solution is to separate the graph into two layers: one layer with positive links, the other with only negative links 2. In general, we would like to have relatively many positive links within communities, while for negative links the opposite holds: we want many negative links between communities. We can easily do this within the multiplex layer framework by passing in a negative layer weight. For example, suppose we have a graph G with possibly negative weights. We can then separate it into a positive and negative graph as follows:

>>> G_pos = G.subgraph_edges(G.es.select(weight_gt = 0), delete_vertices=False);
>>> G_neg = G.subgraph_edges(G.es.select(weight_lt = 0), delete_vertices=False);
>>> G_neg.es['weight'] = [-w for w in G_neg.es['weight']];

We can then simply detect communities using;

>>> part_pos = louvain.ModularityVertexPartition(G_pos, weights='weight');
>>> part_neg = louvain.ModularityVertexPartition(G_neg, weights='weight');
>>> diff = optimiser.optimise_partition_multiplex(
...   [part_pos, part_neg],
...   layer_weights=[1,-1]);

Bipartite

For some methods it may be possible to to community detection in bipartite networks. Bipartite networks are special in the sense that they have only links between the two different classes, and no links within a class are allowed. For example, there might be products and customers, and there is a link between \(i\) and \(j\) if a product \(i\) is bought by a customer \(j\). In this case, there are no links among products, nor among customers. One possible approach is simply project this bipartite network into the one or the other class and then detect communities. But then the correspondence between the communities in the two different projections is lost. Detecting communities in the bipartite network can therefore be useful.

Setting this up requires a bit of a creative approach, which is why it is also explicitly explained here. We will explain it for the CPM method, and then show how this works the same for some related measures. In the case of CPM you would like to be able to set three different resolution parameters: one for within each class \(\gamma_0, \gamma_1\), and one for the links between classes, \(\gamma_{01}\). Then the formulation would be

\[Q = \sum_{ij} [A_{ij} - (\gamma_0\delta(s_i,0) + \gamma_1\delta(s_i,1)) \delta(s_i,s_j) - \gamma_{01}(1 - \delta(s_i, s_j)) ]\delta(\sigma_i, \sigma_j)\]

where \(s_i\) denotes the bipartite class of a node and \(\sigma_i\) the community of the node as elsewhere in the documentation. Rewriting as a sum over communities gives a bit more insight

\[Q = \sum_c (e_c - \gamma_{01} 2 n_c(0) n_c(1) - \gamma_0 n^2_c(0) - \gamma_1 n^2_c(1))\]

where \(n_c(0)\) is the number of nodes in community \(c\) of class 0 (and similarly for 1) and \(e_c\) is the number of edges within community \(c\). We denote by \(n_c = n_c(0) + n_c(1)\) the total number of nodes in community \(c\). Note that

\[\begin{split}n_c^2 &= (n_c(0) + n_c(1))^2 \\ &= n_c(0)^2 + 2 n_c(0) n_c(1) + n_c(1)^2\end{split}\]

We then create three different layers: (1) all nodes have node_size = 1 and all relevant links; (2) only nodes of class 0 have node_size = 1 and no links; (3) only nodes of class 1 have node_size = 1 and no links. If we add the first with resolution parameter \(\gamma_{01}\), and the others with resolution parameters \(\gamma_{01} - \gamma_0\) and \(\gamma_{01} - \gamma_1\), but the latter two with a layer weight of -1 while the first layer has layer weight 1, we obtain the following:

\[\begin{split}Q &= \sum_c (e_c - \gamma_{01} n_c^2) -\sum_c (- (\gamma_{01} - \gamma_0) n_c(0)^2) -\sum_c (- (\gamma_{01} - \gamma_1) n_c(0)^2) \\ &= \sum_c [e_c - \gamma_{01} 2 n_c(0) n_c(1) - \gamma_{01} n_c(0)^2 - \gamma_{01} n_c(1)^2) + ( \gamma_{01} - \gamma_0) n_c(0)^2 + ( \gamma_{01} - \gamma_1) n_c(1)^2 ] \\ &= \sum_c (e_c - \gamma_{01} 2 n_c(0) n_c(1) - \gamma_{0} n_c(0)^2 - \gamma_{1} n_c(1)^2) \\\end{split}\]

Hence detecting communities with these three layers corresponds to detecting communities in bipartite networks. Although we worked out this example for directed network including self-loops (since it is easiest), it works out similarly for undirected networks (with or without self-loops). This only corresponds to the CPM method. However, using a little additional trick, we can also make this work for modularity. Essentially, modularity is nothing else than CPM with the node_size set to the degree, and the resolution parameter set to \(\gamma = \frac{1}{2m}\). In particular, in general (i.e. not specifically for bipartite graph) if node_sizes=G.degree() we then obtain

\[Q = \sum_{ij} A_{ij} - \gamma k_i k_j\]

In the case of bipartite graphs something similar is obtained, but then correctly adapted (as long as the resolution parameter is also appropriately rescaled). Note that this is only possible for modularity for undirected graphs. Hence, we can also detect communities in bipartite networks using modularity by using this little trick.

All of this has been implemented in the constructor Bipartite(). You can simply pass in a bipartite network with the classes appropriately defined in G.vs['type'] or equivalent. This function assumes the two classes are coded by 0 and 1, and if this is not the case it will try to convert it into such categories by ig.UniqueIdGenerator().

An explicit example of this:

>>> p_01, p_0, p_1 = louvain.CPMVertexPartition.Bipartite(G,
...                    resolution_parameter_01=0.1);
>>> diff = optimiser.optimise_partition_multiplex([p_01, p_0, p_1],
...                                        layer_weights=[1, -1, -1]);

Slices to layers

The multiplex formulation as layers has two limitations: (1) each graph needs to have an identical vertex set; (2) each node is only in a single community. Ideally, one would like to relax both these requirements, so that you can work with graphs that do not need to have identical nodes and where nodes can be in different communities in different layers. For example, a person could be in one community when looking at his professional relations, but in another community looking at his personal relations. Perhaps more commonly: a person could be in one community at time 1 and in another community at time 2.

Fortunately, this is also possible with this package. We call the more general formulation slices in contrast to the layers required by the earlier functions. Slices are then just different graphs, which do not need to have the same vertex set in any way. The idea is to build one big graph out of all the slices and then decompose it again in layers that correspond with slices. The key element is that some slices are coupled: for example two consecutive time windows, or simply two different slices of types of relations. Because any two slices can be coupled in theory, we represent the coupling itself again with a graph. The nodes of this coupling graph thus are slices, and the (possibly weighted) links in the coupling graph represent the (possibly weighted) couplings between slices. Below an example with three different time slices, where slice 1 is coupled to slice 2, which in turn is coupled to slice 3:

The coupling graph thus consists of three nodes and a simple line structure: 1 -- 2 -- 3. We convert this into layers by putting all nodes of all slices in one big network. Each node is thus represented by a tuple (node, slice) in a certain sense. Out of this big network, we then only take those edges that are defined between nodes of the same slice, which then constitutes a single layer. Finally, we need one more layer for the couplings. In addition, for methods such as CPMVertexPartition, so-called node_sizes are required, and for them to properly function, they should be set to 0 (which is handled appropriately by the package). We thus obtain equally many layers as we have slices, and we need one more layer for representing the interslice couplings. For the example provided above, we thus obtain the following:

To transform slices into layers using a coupling graph, this package provides layers_to_slices(). For the example above, this would function as follows. First create the coupling graph assuming we have three slices G_1, G_2 and G_3:

>>> G_coupling = ig.Graph.Formula('1 -- 2 -- 3');
>>> G_coupling.es['weight'] = 0.1; # Interslice coupling strength
>>> G_coupling.vs['slice'] = [G_1, G_2, G_3]

Then we convert them to layers

>>> layers, interslice_layer, G_full = louvain.slices_to_layers(G_coupling);

Now we still have to create partitions for all the layers. We can freely choose here to use the same partition types for all partitions, or to use different types for different layers.

Warning

The interslice layer should usually be of type CPMVertexPartition with a resolution_parameter=0 and node_sizes set to 0. The G.vs[node_size] is automatically set to 0 for all nodes in the interslice layer in slices_to_layers(), so you can simply pass in the attribute node_size. Unless you know what you are doing, simply use these settings.

Warning

When using methods that accept a node_size argument, this should always be used. This is the case for CPMVertexPartition, RBERVertexPartition, SurpriseVertexPartition and SignificanceVertexPartition.

>>> partitions = [louvain.CPMVertexPartition(H, node_sizes='node_size',
...                                          weights='weight', resolution_parameter=gamma)
...               for H in layers];
>>> interslice_partition = louvain.CPMVertexPartition(interslice_layer, resolution_parameter=0,
...                                                   node_sizes='node_size', weights='weight');

You can then simply optimise these partitions as before using optimise_partition_multiplex():

>>> diff = optimiser.optimise_partition_multiplex(partitions + [interslice_partition]);

Temporal community detection

One of the most common tasks for converting slices to layers is that we have slices at different points in time. We call this temporal community detection. Because it is such a common task, we provide several helper functions to simplify the above process. Let us assume again that we have three slices G_1, G_2 and G_3 as in the example above. The most straightforward function is find_partition_temporal():

>>> membership, improvement = louvain.find_partition_temporal(
...                             [G_1, G_2, G_3],
...                             louvain.CPMVertexPartition,
...                             interslice_weight=0.1,
...                             resolution_parameter=gamma)

This function only returns the membership vectors for the different time slices, rather than actual partitions.

Rather than directly detecting communities, you can also obtain the actual partitions in a slightly more convenient way using time_slices_to_layers():

>>> layers, interslice_layer, G_full = \
...               louvain.time_slices_to_layers([G_1, G_2, G_3],
...                                             interslice_weight=0.1);
>>> partitions = [louvain.CPMVertexPartition(H, node_sizes='node_size',
...                                          weights='weight',
...                                          resolution_parameter=gamma)
...               for H in layers];
>>> interslice_partition = \
...               louvain.CPMVertexPartition(interslice_layer, resolution_parameter=0,
...                                          node_sizes='node_size', weights='weight');
>>> diff = optimiser.optimise_partition_multiplex(partitions + [interslice_partition]);

Both these functions assume that the interslice coupling is always identical for all slices. If you want more finegrained control, you will have to use the earlier explained functions.

References

1

Mucha, P. J., Richardson, T., Macon, K., Porter, M. A., & Onnela, J.-P. (2010). Community structure in time-dependent, multiscale, and multiplex networks. Science, 328(5980), 876–8. 10.1126/science.1184819

2

Traag, V. A., & Bruggeman, J. (2009). Community detection in networks with positive and negative links. Physical Review E, 80(3), 036115. 10.1103/PhysRevE.80.036115

Implementation

If you have a cool new idea for a better method, and you want to optimise it, you can easily plug it in the current tool. This section explains how the package is setup internally, and how you can extend it. Most of this concerns C++, and python only comes in when exposing the resulting classes.

Method

All methods in the end derive from MutableVertexPartition, which implements almost all necessary details, such as moving actual nodes while maintaining the internal administration. Similarly, it provides all the necessary functionality for initialising a partition. Additionally, there are two abstract classes that derive from this base class: ResolutionParameterVertexPartition and LinearResolutionParameterVertexPartition (which in turn derives from the former class). If you want a method with a resolution parameter, you should derive from one of these two classes, otherwise, simply from the base class MutableVertexPartition.

There are two functions that you need to implement yourself: diff_move() and quality(). Note that they should always be consistent, so that we can double check the internal consistency. You should also ensure that the diff_move function can be correctly used on any aggregate graph (i.e. moving a node in the aggregate graph indeed corresponds to moving a set of nodes in the individual graph).

That’s it. In principle, you could now use and test the method in C++.

Python

Exposing the method to python takes a bit more effort. There are various places in which you need to change/add things. In the following, we assume you created a new class called CoolVertexPartition. In order of dependencies, it goes as follows:

1. Your own new VertexPartition class should add some specific methods. In particular, you need to ensure you create a method

   CoolVertexPartition* CoolVertexPartition::create(Graph* graph)
   {
     return new CoolVertexPartition(graph);
   }
   

   and

   CoolVertexPartition* CoolVertexPartition::create(Graph* graph, vector<size_t> const& membership)
   {
     return new CoolVertexPartition(graph, membership);
   }
   

   These methods ensure that based on a current partition, we can create a new partition (without knowing its type).

2. In python_partition_interface.cpp some methods need to be added. In particular

   PyObject* _new_CoolVertexPartition(PyObject *self, PyObject *args, PyObject *keywds)
   

   You should be able to simply copy an existing method, and adapt it to your own needs.

3. These methods need to be exposed in pynterface.h. In particular, you need to add the method you created in step (2) to louvain_funcs[]. Again, you should be able to simply copy an existing line.

4. You can then finally create the Python class in VertexPartition.py. The base class derives from the VertexClustering from igraph, so that it is compatible with all operations in igraph. You should add the method as follows:

   class CoolVertexPartition(MutableVertexPartition):
   
     def __init__(self, ... ):
       ...
   

   Again, you should be able to copy the outline for another class and adapt it to your own needs. Don’t forget to change to docstring to update the documentation so that everybody knows how your new cool method works.

5. Expose your newly created python class directly in __init__.py by importing it:

   from .VertexPartition import CoolVertexPartition
   

That’s it! You’re done and should now be able to find communities using your new CoolVertexPartition:

>>> louvain.find_partition(G, louvain.CoolVertexPartition); 

Reference

Reference

Module functions

This package implements the louvain algorithm in C++ and exposes it to python. It relies on (python-)igraph for it to function. Besides the relative flexibility of the implementation, it also scales well, and can be run on graphs of millions of nodes (as long as they can fit in memory). Each method is represented by a different class, all of whom derive from MutableVertexPartition. In addition, multiplex graphs are supported as layers, which also supports multislice representations.

Examples

The simplest example just finds a partition using modularity

>>> G = ig.Graph.Tree(100, 3)
>>> partition = louvain.find_partition(G, louvain.ModularityVertexPartition)

Alternatively, one can access the different optimisation routines individually and construct partitions oneself. These partitions can then be optimised by constructing an Optimiser object and running optimise_partition().

>>> G = ig.Graph.Tree(100, 3)
>>> partition = louvain.CPMVertexPartition(G, resolution_parameter = 0.1)
>>> optimiser = louvain.Optimiser()
>>> diff = optimiser.optimise_partition(partition)

The Optimiser class contains also the different subroutines that are used internally by optimise_partition(). In addition, through the Optimiser class there are various options available for changing some of the optimisation procedure which can affect both speed and quality, which are not immediately available in louvain.find_partition().

louvain.find_partition(graph, partition_type, initial_membership=None, weights=None, seed=None, **kwargs)

Detect communities using the default settings.

This function detects communities given the specified method in the partition_type. This should be type derived from VertexPartition.MutableVertexPartition, e.g. ModularityVertexPartition or CPMVertexPartition. Optionally an initial membership and edge weights can be provided. Remaining **kwargs are passed on to the constructor of the partition_type, including for example a resolution_parameter.

Parameters

• graph (ig.Graph) – The graph for which to detect communities.

• partition_type (type of :class:`) – The type of partition to use for optimisation.

• initial_membership (list of int) – Initial membership for the partition. If None then defaults to a singleton partition.

• weights (list of double, or edge attribute) – Weights of edges. Can be either an iterable or an edge attribute.

• seed (int) – Seed for the random number generator. By default uses a random seed if nothing is specified.

• **kwargs – Remaining keyword arguments, passed on to constructor of partition_type.

Returns

The optimised partition.

Return type

partition

See also

Optimiser.optimise_partition()

Examples

>>> G = ig.Graph.Famous('Zachary')
>>> partition = louvain.find_partition(G, louvain.ModularityVertexPartition)

louvain.find_partition_multiplex(graphs, partition_type, seed=None, **kwargs)

Detect communities for multiplex graphs.

Each graph should be defined on the same set of vertices, only the edges may differ for different graphs. See Optimiser.optimise_partition_multiplex() for a more detailed explanation.

Parameters

• graphs (list of ig.Graph) – List of louvain.VertexPartition layers to optimise.

• partition_type (type of MutableVertexPartition) – The type of partition to use for optimisation (identical for all graphs).

• seed (int) – Seed for the random number generator. By default uses a random seed if nothing is specified.

• **kwargs – Remaining keyword arguments, passed on to constructor of partition_type.

Returns

• list of int – membership of nodes.

• float – Improvement in quality of combined partitions, see Optimiser.optimise_partition_multiplex().

Notes

We don’t return a partition in this case because a partition is always defined on a single graph. We therefore simply return the membership (which is the same for all layers).

See also

Optimiser.optimise_partition_multiplex(), slices_to_layers()

Examples

>>> n = 100
>>> G_1 = ig.Graph.Lattice([n], 1)
>>> G_2 = ig.Graph.Lattice([n], 1)
>>> membership, improvement = louvain.find_partition_multiplex([G_1, G_2],
...                                                            louvain.ModularityVertexPartition)

louvain.find_partition_temporal(graphs, partition_type, interslice_weight=1, slice_attr='slice', vertex_id_attr='id', edge_type_attr='type', weight_attr='weight', seed=None, **kwargs)

Detect communities for temporal graphs.

Each graph is considered to represent a time slice and does not necessarily need to be defined on the same set of vertices. Nodes in two consecutive slices are identified on the basis of the vertex_id_attr, i.e. if two nodes in two consecutive slices have an identical value of the vertex_id_attr they are coupled. The vertex_id_attr should hence be unique in each slice. The nodes are then coupled with a weight of interslice_weight which is set in the edge attribute weight_attr. No weight is set if the interslice_weight is None (i.e. corresponding in practice with a weight of 1). See time_slices_to_layers() for a more detailed explanation.

Parameters

• graphs (list of ig.Graph) – List of louvain.VertexPartition layers to optimise.

• partition_type (type of VertexPartition.MutableVertexPartition) – The type of partition to use for optimisation (identical for all graphs).

• interslice_weight (float) – The weight of the coupling between two consecutive time slices.

• slice_attr (string) – The vertex attribute to use for indicating the slice of a node.

• vertex_id_attr (string) – The vertex to use to identify nodes.

• edge_type_attr (string) – The edge attribute to use for indicating the type of link (interslice or intraslice).

• weight_attr (string) – The edge attribute used to indicate the weight.

• seed (int) – Seed for the random number generator. By default uses a random seed if nothing is specified.

• **kwargs – Remaining keyword arguments, passed on to constructor of partition_type.

Returns

• list of membership – list containing for each slice the membership vector.

• float – Improvement in quality of combined partitions, see Optimiser.optimise_partition_multiplex().

See also

time_slices_to_layers(), slices_to_layers()

Examples

>>> n = 100
>>> G_1 = ig.Graph.Lattice([n], 1)
>>> G_1.vs['id'] = range(n)
>>> G_2 = ig.Graph.Lattice([n], 1)
>>> G_2.vs['id'] = range(n)
>>> membership, improvement = louvain.find_partition_temporal([G_1, G_2],
...                                                           louvain.ModularityVertexPartition,
...                                                           interslice_weight=1)

louvain.slices_to_layers(G_coupling, slice_attr='slice', vertex_id_attr='id', edge_type_attr='type', weight_attr='weight')

Convert a coupling graph of slices to layers of graphs.

This function converts a graph of slices to layers so that they can be used with this package. This function assumes that the slices are represented by nodes in G_coupling, and stored in the attribute slice_attr. In other words, G_coupling.vs[slice_attr] should contain ig.Graph s . The slices will be converted to layers, and nodes in different slices will be coupled if the two slices are connected in G_coupling. Nodes in two connected slices are identified on the basis of the vertex_id_attr, i.e. if two nodes in two connected slices have an identical value of the vertex_id_attr they will be coupled. The vertex_id_attr should hence be unique in each slice. The weight of the coupling is determined by the weight of this link in G_coupling, as determined by the weight_attr.

Parameters

• G_coupling (ig.Graph) – The graph connecting the different slices.

• slice_attr (string) – The vertex attribute which contains the slices.

• vertex_id_attr (string) – The vertex attribute which is used to identify whether two nodes in two slices represent the same node, and hence, should be coupled.

• edge_type_attr (string) – The edge attribute to use for indicating the type of link (interslice or intraslice).

• weight_attr (string) – The edge attribute used to indicate the (coupling) weight.

Returns

• G_layers (list of ig.Graph) – A list of slices converted to layers.

• G_interslice (ig.Graph) – The interslice coupling layer.

• G (ig.Graph) – The complete graph containing all layers and interslice couplings.

Notes

The distinction between slices and layers is not easy to grasp. Slices in this context refer to graphs that somehow represents different aspects of a network. The simplest example is probably slices that represents time: there are different snapshots network across time, and each snapshot is considered a slice. Some nodes may drop out of the network over time, while others enter the network. Edges may change over time, or the weight of the links may change over time. This is just the simplest example of a slice, and there may be different, more complex possibilities. Below an example with three time slices:

Now in order to optimise partitions across these different slices, we represent them slightly differently, namely as layers. The idea of layers is that all graphs always are defined on the same set of nodes, and that only the links differ for different layers. We thus create new nodes as combinations of original nodes and slices. For example, if node 1 existed in both slice 1 and in slice 2, we will thus create two nodes to build the layers: a node 1-1 and a node 1-2. Additionally, if the slices are connected in the slice graph, the two nodes would also be connected, so there would be a linke between node 1-1 and 1-2. Different slices will then correspond to different layers: each layer only contains the link for that particular slice. In addition, for methods such as CPMVertexPartition, so-called node_sizes are required, and for them to properly function, they should be set to 0 (which is handled appropriately in this function, and stored in the vertex attribute node_size). We thus obtain equally many layers as we have slices, and we need one more layer for representing the interslice couplings. For the example provided above, we thus obtain the following:

The idea of doing community detection with slices is further detailed in [1].

References

1

Mucha, P. J., Richardson, T., Macon, K., Porter, M. A., & Onnela, J.-P. (2010). Community structure in time-dependent, multiscale, and multiplex networks. Science, 328(5980), 876-8. 10.1126/science.1184819

See also

find_partition_temporal(), time_slices_to_layers()

louvain.time_slices_to_layers(graphs, interslice_weight=1, slice_attr='slice', vertex_id_attr='id', edge_type_attr='type', weight_attr='weight')

Convert time slices to layer graphs.

Each graph is considered to represent a time slice. This function simply connects all the consecutive slices (i.e. the slice graph) with an interslice_weight. The further conversion is then delegated to slices_to_layers(), which also provides further details.

See also

find_partition_temporal(), slices_to_layers()

Optimiser

class louvain.Optimiser

Bases: object

Class for doing community detection using the Louvain algorithm.

The optimiser class provides a number of different methods for optimising a given partition. The overall optimise procedure optimise_partition() calls move_nodes() then aggregates the graph and repeats the same procedure. For calculating the actual improvement of moving a node (corresponding a subset of nodes in the aggregate graph), the code relies on diff_move() which provides different values for different methods (e.g. modularity or CPM). Finally, the Optimiser class provides a routine to construct a resolution_profile() on a resolution parameter.

Create a new Optimiser object

property consider_comms

Determine how alternative communities are considered for moving a node for optimising a partition.

Nodes will only move to alternative communities that improve the given quality function.

Notes

This attribute should be set to one of the following values

• louvain.ALL_NEIGH_COMMS Consider all neighbouring communities for moving.

• louvain.ALL_COMMS Consider all communities for moving. This is especially useful in the case of negative links, in which case it may be better to move a node to a non-neighbouring community.

• louvain.RAND_NEIGH_COMM Consider a random neighbour community for moving. The probability to choose a community is proportional to the number of neighbours a node has in that community.

• louvain.RAND_COMM Consider a random community for moving. The probability to choose a community is proportional to the number of nodes in that community.

property consider_empty_community

if True consider also moving nodes to an empty community (default).

Type

boolean

move_nodes(partition, consider_comms=None)

Move nodes to alternative communities for optimising the partition.

Parameters

• partition – The partition for which to move nodes.

• consider_comms – If None uses consider_comms, but can be set to something else.

Returns

Improvement in quality function.

Return type

float

Notes

When moving nodes, the function loops over nodes and considers moving the node to an alternative community. Which community depends on consider_comms. The function terminates when no more nodes can be moved to an alternative community.

Examples

>>> G = ig.Graph.Famous('Zachary')
>>> optimiser = louvain.Optimiser()
>>> partition = louvain.ModularityVertexPartition(G)
>>> diff = optimiser.move_nodes(partition)

optimise_partition(partition)

Optimise the given partition.

Parameters

partition – The MutableVertexPartition to optimise.

Returns

Improvement in quality function.

Return type

float

Examples

>>> G = ig.Graph.Famous('Zachary')
>>> optimiser = louvain.Optimiser()
>>> partition = louvain.ModularityVertexPartition(G)
>>> diff = optimiser.optimise_partition(partition)

optimise_partition_multiplex(partitions, layer_weights=None)

Optimise the given partitions simultaneously.

Parameters

• partitions – List of MutableVertexPartition layers to optimise.

• layer_weights – List of weights of layers.

Returns

Improvement in quality of combined partitions, see Notes.

Return type

float

Notes

This method assumes that the partitions are defined for graphs with the same vertices. The connections between the vertices may be different, but the vertices themselves should be identical. In other words, all vertices should have identical indices in all graphs (i.e. node i is assumed to be the same node in all graphs). The quality of the overall partition is simply the sum of the individual qualities for the various partitions, weighted by the layer_weight. If we denote by \(Q_k\) the quality of layer \(k\) and the weight by \(\lambda_k\), the overall quality is then

\[Q = \sum_k \lambda_k Q_k.\]

This is particularly useful for graphs containing negative links. When separating the graph in two graphs, the one containing only the positive links, and the other only the negative link, by supplying a negative weight to the latter layer, we try to find relatively many positive links within a community and relatively many negative links between communities. Note that in this case it may be better to assign a node to a community to which it is not connected so that consider_comms may be better set to louvain.ALL_COMMS.

Besides multiplex graphs where each node is assumed to have a single community, it is also useful in the case of for example multiple time slices, or in situations where nodes can have different communities in different slices. The package includes some special helper functions for using optimise_partition_multiplex() in such cases, where there is a conversion required from (time) slices to layers suitable for use in this function.

See also

slices_to_layers(), time_slices_to_layers(), find_partition_multiplex(), find_partition_temporal()

Examples

>>> G_pos = ig.Graph.SBM(100, pref_matrix=[[0.5, 0.1], [0.1, 0.5]], block_sizes=[50, 50])
>>> G_neg = ig.Graph.SBM(100, pref_matrix=[[0.1, 0.5], [0.5, 0.1]], block_sizes=[50, 50])
>>> optimiser = louvain.Optimiser()
>>> partition_pos = louvain.ModularityVertexPartition(G_pos)
>>> partition_neg = louvain.ModularityVertexPartition(G_neg)
>>> diff = optimiser.optimise_partition_multiplex(
...                     partitions=[partition_pos, partition_neg],
...                     layer_weights=[1,-1])

resolution_profile(graph, partition_type, resolution_range, weights=None, bisect_func=<function Optimiser.<lambda>>, min_diff_bisect_value=1, min_diff_resolution=0.001, linear_bisection=False, number_iterations=1, **kwargs)

Use bisectioning on the resolution parameter in order to construct a resolution profile.

Parameters

• graph – The graph for which to construct a resolution profile.

• partition_type – The type of MutableVertexPartition used to find a partition (must support resolution parameters obviously).

• resolution_range – The range of resolution values that we would like to scan.

• weights – If provided, indicates the edge attribute to use as a weight.

• bisect_func – The function used for bisectioning. For the methods currently implemented, this should usually not be altered.

• min_diff_bisect_value – The difference in the value returned by the bisect_func below which the bisectioning stops (i.e. by default, a difference of a single edge does not trigger further bisectioning).

• min_diff_resolution – The difference in resolution below which the bisectioning stops. For positive differences, the logarithmic difference is used by default, i.e. diff = log(res_1) - log(res_2) = log(res_1/res_2), for which diff > min_diff_resolution to continue bisectioning. Set the linear_bisection to true in order to use only linear bisectioning (in the case of negative resolution parameters for example, which can happen with negative weights).

• linear_bisection – Whether the bisectioning will be done on a linear or on a logarithmic basis (if possible).

• number_iterations – Indicates the number of iterations of the algorithm to run. If negative (or zero) the algorithm is run until a stable iteration.

Returns

A list of partitions for different resolutions.

Return type

list of MutableVertexPartition

Examples

>>> G = ig.Graph.Famous('Zachary')
>>> optimiser = louvain.Optimiser()
>>> profile = optimiser.resolution_profile(G, louvain.CPMVertexPartition,
...                                        resolution_range=(0,1))

set_rng_seed(value)

Set the random seed for the random number generator.

Parameters

value – The integer seed used in the random number generator

MutableVertexPartition

class louvain.VertexPartition.MutableVertexPartition(graph, initial_membership=None)

Bases: VertexClustering

Contains a partition of graph, derives from ig.VertexClustering.

This class contains the basic implementation for optimising a partition. Specifically, it implements all the administration necessary to keep track of the partition from various points of view. Internally, it keeps track of the number of internal edges (or total weight), the size of the communities, the total incoming degree (or weight) for a community, et cetera.

In order to keep the administration up-to-date, all changes in a partition should be done through move_node() or set_membership(). The first moves a node from one community to another, and updates the administration. The latter simply updates the membership vector and updates the administration.

The basic idea is that diff_move() computes the difference in the quality function if we would call move_node() for the same move. These functions are overridden in any derived classes to provide an actual implementation. These functions are used by Optimiser to optimise the partition.

Warning

This base class should never be used in practice, since only derived classes provide an actual implementation.

Parameters

• graph – The ig.Graph on which this partition is defined.

• membership – The membership vector of this partition. Membership[i] = c implies that node i is in community c. If None, it is initialised with a singleton partition community, i.e. membership[i] = i.

classmethod FromPartition(partition, **kwargs)

Create a new partition from an existing partition.

Parameters

• partition – The MutableVertexPartition to replicate.

• **kwargs – Any remaining keyword arguments will be passed on to the constructor of the new partition.

Notes

This may for example come in handy when determining the quality of a partition using a different method. Suppose that we already have a partition p and that we want to determine the Significance of that partition. We can then simply use

>>> p = louvain.find_partition(ig.Graph.Famous('Zachary'),
...                            louvain.ModularityVertexPartition)
>>> sig = louvain.SignificanceVertexPartition.FromPartition(p).quality()

aggregate_partition(membership_partition=None)

Aggregate the graph according to the current partition and provide a default partition for it.

The aggregated graph can then be found as a parameter of the partition partition.graph.

Notes

This function contrasts to the function cluster_graph in igraph itself, which also provides the aggregate graph, but we may require setting the appropriate resolution_parameter, weights and node_sizes. In particular, this function also ensures that the quality defined on the aggregate partition is identical to the quality defined on the original partition.

Examples

>>> G = ig.Graph.Famous('Zachary')
>>> partition = louvain.find_partition(G, louvain.ModularityVertexPartition)
>>> aggregate_partition = partition.aggregate_partition(partition)
>>> aggregate_graph = aggregate_partition.graph
>>> aggregate_partition.quality() == partition.quality()
True

diff_move(v, new_comm)

Calculate the difference in the quality function if node v is moved to community new_comm.

Parameters

• v – The node to move.

• new_comm – The community to move to.

Returns

Difference in quality function.

Return type

float

Notes

The difference returned by diff_move should be equivalent to first determining the quality of the partition, then calling move_node, and then determining again the quality of the partition and looking at the difference. In other words

>>> partition = louvain.find_partition(ig.Graph.Famous('Zachary'),
...                            louvain.ModularityVertexPartition)
>>> diff = partition.diff_move(v=0, new_comm=0)
>>> q1 = partition.quality()
>>> partition.move_node(v=0, new_comm=0)
>>> q2 = partition.quality()
>>> round(diff, 10) == round(q2 - q1, 10)
True

Warning

Only derived classes provide actual implementations, the base class provides no implementation for this function.

from_coarse_partition(partition, coarse_node=None)

Update current partition according to coarser partition.

Parameters

• partition (MutableVertexPartition) – The coarser partition used to update the current partition.

• coarse_node (list of int) – The coarser node which represent the current node in the partition.

Notes

This function is to be used to determine the correct partition for an aggregated graph. In particular, suppose we move nodes and then get an aggregate graph.

>>> diff = optimiser.move_nodes(partition)
>>> aggregate_partition = partition.aggregate_partition()

Now we also move nodes in the aggregate partition

>>> diff = optimiser.move_nodes(aggregate_partition)

Now we improved the quality function of aggregate_partition, but this is not yet reflected in the original partition. We can thus call

>>> partition.from_coarse_partition(aggregate_partition)

so that partition now reflects the changes made to aggregate_partition.

The coarse_node can be used it the aggregate_partition is not defined based on the membership of this partition. In particular the membership of this partition is defined as follows:

>>> for v in G.vs:
...   partition.membership[v] = aggregate_partition.membership[coarse_node[v]] 

If coarse_node is None it is assumed the coarse node was defined based on the membership of the current partition, so that

>>> for v in G.vs:
...   partition.membership[v] = aggregate_partition.membership[partition.membership[v]] 

This can be useful when the aggregate partition is defined on a more refined partition.

move_node(v, new_comm)

Move node v to community new_comm.

Parameters

• v – Node to move.

• new_comm – Community to move to.

Examples

>>> G = ig.Graph.Famous('Zachary')
>>> partition = louvain.ModularityVertexPartition(G)
>>> partition.move_node(0, 1)

quality()

The current quality of the partition.

renumber_communities()

Renumber the communities so that they are numbered in decreasing size.

Notes

The sort is not necessarily stable.

set_membership(membership)

Set membership.

total_possible_edges_in_all_comms()

The total possible number of edges in all communities.

Notes

If we denote by \(n_c\) the number of nodes in community \(c\), this is simply

\[\sum_c \binom{n_c}{2}\]

total_weight_from_comm(comm)

The total weight (i.e. number of edges) from a community.

Parameters

comm – Community

Notes

This includes all edges, also the ones that are internal to a community. Sometimes this is also referred to as the community (out)degree.

See also

total_weight_to_comm(), total_weight_in_comm(), total_weight_in_all_comms()

total_weight_in_all_comms()

The total weight (i.e. number of edges) within all communities.

Notes

This should be equal to simply the sum of total_weight_in_comm for all communities.

See also

total_weight_to_comm(), total_weight_from_comm(), total_weight_in_comm()

total_weight_in_comm(comm)

The total weight (i.e. number of edges) within a community.

Parameters

comm – Community

See also

total_weight_to_comm(), total_weight_from_comm(), total_weight_in_all_comms()

total_weight_to_comm(comm)

The total weight (i.e. number of edges) to a community.

Parameters

comm – Community

Notes

This includes all edges, also the ones that are internal to a community. Sometimes this is also referred to as the community (in)degree.

See also

total_weight_from_comm(), total_weight_in_comm(), total_weight_in_all_comms()

weight_from_comm(v, comm)

The total number of edges (or sum of weights) to node v from community comm.

See also

weight_to_comm()

weight_to_comm(v, comm)

The total number of edges (or sum of weights) from node v to community comm.

See also

weight_from_comm()

ModularityVertexPartition

class louvain.ModularityVertexPartition(graph, initial_membership=None, weights=None)

Bases: MutableVertexPartition

Implements modularity.

Notes

The quality function is

\[Q = \frac{1}{2m} \sum_{ij} \left(A_{ij} - \frac{k_i k_j}{2m} \right)\delta(\sigma_i, \sigma_j)\]

where \(A\) is the adjacency matrix, \(k_i\) is the degree of node \(i\), \(m\) is the total number of edges, \(\sigma_i\) denotes the community of node \(i\) and \(\delta(\sigma_i, \sigma_j) = 1\) if \(\sigma_i = \sigma_j\) and 0 otherwise.

This can alternatively be formulated as a sum over communities:

\[Q = \frac{1}{2m} \sum_{c} \left(m_c - \frac{K_c^2}{4m} \right)\]

where \(m_c\) is the number of internal edges of community \(c\) and \(K_c = \sum_{i \mid \sigma_i = c} k_i\) is the total degree of nodes in community \(c\).

References

1

Newman, M. E. J., & Girvan, M. (2004). Finding and evaluating community structure in networks. Physical Review E, 69(2), 026113. 10.1103/PhysRevE.69.026113

Parameters

• graph (ig.Graph) – Graph to define the partition on.

• initial_membership (list of int) – Initial membership for the partition. If None then defaults to a singleton partition.

• weights (list of double, or edge attribute) – Weights of edges. Can be either an iterable or an edge attribute.

RBConfigurationVertexPartition

class louvain.RBConfigurationVertexPartition(graph, initial_membership=None, weights=None, resolution_parameter=1.0)

Bases: LinearResolutionParameterVertexPartition

Implements Reichardt and Bornholdt’s Potts model with a configuration null model.

This quality function uses a linear resolution parameter.

Notes

The quality function is

\[Q = \sum_{ij} \left(A_{ij} - \gamma \frac{k_i k_j}{2m} \right)\delta(\sigma_i, \sigma_j)\]

where \(A\) is the adjacency matrix, \(k_i\) is the degree of node \(i\), \(m\) is the total number of edges, \(\sigma_i\) denotes the community of node \(i\), \(\delta(\sigma_i, \sigma_j) = 1\) if \(\sigma_i = \sigma_j\) and 0 otherwise, and, finally \(\gamma\) is a resolution parameter.

This can alternatively be formulated as a sum over communities:

\[Q = \sum_{c} \left(m_c - \gamma\frac{K_c^2}{4m} \right)\]

where \(m_c\) is the number of internal edges of community \(c\) and \(K_c = \sum_{i \mid \sigma_i = c} k_i\) is the total degree of nodes in community \(c\).

Note that this is the same as ModularityVertexPartition for \(\gamma=1\) and using the normalisation by \(2m\).

References

1

Reichardt, J., & Bornholdt, S. (2006). Statistical mechanics of community detection. Physical Review E, 74(1), 016110. 10.1103/PhysRevE.74.016110

Parameters

• graph (ig.Graph) – Graph to define the partition on.

• initial_membership (list of int) – Initial membership for the partition. If None then defaults to a singleton partition.

• weights (list of double, or edge attribute) – Weights of edges. Can be either an iterable or an edge attribute.

• resolution_parameter (double) – Resolution parameter.

property resolution_parameter

Resolution parameter.

RBERVertexPartition

class louvain.RBERVertexPartition(graph, initial_membership=None, weights=None, node_sizes=None, resolution_parameter=1.0)

Bases: LinearResolutionParameterVertexPartition

Implements Reichardt and Bornholdt’s Potts model with a configuration null model.

This quality function uses a linear resolution parameter.

Notes

The quality function is

\[Q = \sum_{ij} \left(A_{ij} - \gamma p \right)\delta(\sigma_i, \sigma_j)\]

where \(A\) is the adjacency matrix,

\[p = \frac{m}{\binom{n}{2}}\]

is the overall density of the graph, \(\sigma_i\) denotes the community of node \(i\), \(\delta(\sigma_i, \sigma_j) = 1\) if \(\sigma_i = \sigma_j\) and 0 otherwise, and, finally \(\gamma\) is a resolution parameter.

This can alternatively be formulated as a sum over communities:

\[Q = \sum_{c} \left[m_c - \gamma p \binom{n_c}{2} \right]\]

where \(m_c\) is the number of internal edges of community \(c\) and \(n_c\) the number of nodes in community \(c\).

References

1

Reichardt, J., & Bornholdt, S. (2006). Statistical mechanics of community detection. Physical Review E, 74(1), 016110. 10.1103/PhysRevE.74.016110

Parameters

• graph (ig.Graph) – Graph to define the partition on.

• initial_membership (list of int) – Initial membership for the partition. If None then defaults to a singleton partition.

• weights (list of double, or edge attribute) – Weights of edges. Can be either an iterable or an edge attribute.

• node_sizes (list of int, or vertex attribute) – Sizes of nodes are necessary to know the size of communities in aggregate graphs. Usually this is set to 1 for all nodes, but in specific cases this could be changed.

• resolution_parameter (double) – Resolution parameter.

CPMVertexPartition

class louvain.CPMVertexPartition(graph, initial_membership=None, weights=None, node_sizes=None, resolution_parameter=1.0)

Bases: LinearResolutionParameterVertexPartition

Implements CPM. This quality function uses a linear resolution parameter.

Notes

The quality function is

\[Q = \sum_{ij} \left(A_{ij} - \gamma \right)\delta(\sigma_i, \sigma_j)\]

where \(A\) is the adjacency matrix, \(\sigma_i\) denotes the community of node \(i\), \(\delta(\sigma_i, \sigma_j) = 1\) if \(\sigma_i = \sigma_j\) and 0 otherwise, and, finally \(\gamma\) is a resolution parameter.

This can alternatively be formulated as a sum over communities:

\[Q = \sum_{c} \left[m_c - \gamma \binom{n_c}{2} \right]\]

where \(m_c\) is the number of internal edges of community \(c\) and \(n_c\) the number of nodes in community \(c\).

The resolution parameter \(\gamma\) for this functions has a particularly simple interpretation. The internal density of communities

\[p_c = \frac{m_c}{\binom{n_c}{2}} \geq \gamma\]

is higher than \(\gamma\), while the external density

\[p_{cd} = \frac{m_{cd}}{n_c n_d} \leq \gamma\]

is lower than \(\gamma\). In other words, choosing a particular \(\gamma\) corresponds to choosing to find communities of a particular density, and as such defines communities. Finally, the definition of a community is in a sense independent of the actual graph, which is not the case for any of the other methods (see the reference for more detail).

References

1

Traag, V. A., Van Dooren, P., & Nesterov, Y. (2011). Narrow scope for resolution-limit-free community detection. Physical Review E, 84(1), 016114. 10.1103/PhysRevE.84.016114

Parameters

• graph (ig.Graph) – Graph to define the partition on.

• initial_membership (list of int) – Initial membership for the partition. If None then defaults to a singleton partition.

• weights (list of double, or edge attribute) – Weights of edges. Can be either an iterable or an edge attribute.

• node_sizes (list of int, or vertex attribute) – Sizes of nodes are necessary to know the size of communities in aggregate graphs. Usually this is set to 1 for all nodes, but in specific cases this could be changed.

• resolution_parameter (double) – Resolution parameter.

Bipartite(resolution_parameter_01, resolution_parameter_0=0, resolution_parameter_1=0, degree_as_node_size=False, types='type', **kwargs)

Create three layers for bipartite partitions.

This creates three layers for bipartite partition necessary for detecting communities in bipartite networks. These three layers should be passed to Optimiser.optimise_partition_multiplex() with layer_weights=[1,-1,-1].

Parameters

• graph (ig.Graph) – Graph to define the bipartite partitions on.

• resolution_parameter_01 (double) – Resolution parameter for in between two classes.

• resolution_parameter_0 (double) – Resolution parameter for class 0.

• resolution_parameter_1 (double) – Resolution parameter for class 1.

• degree_as_node_size (boolean) –

  If True use degree as node size instead of 1, to mimic modularity, see Notes.

• types (vertex attribute or list) – Indicator of the class for each vertex. If not 0, 1, it is automatically converted.

• **kwargs – Additional arguments passed on to default constructor of CPMVertexPartition.

• _notes-bipartite (..) –

Notes

For bipartite networks, we would like to be able to set three different resolution parameters: one for within each class \(\gamma_0, \gamma_1\), and one for the links between classes, \(\gamma_{01}\). Then the formulation would be

\[Q = \sum_{ij} [A_{ij} - (\gamma_0\delta(s_i,0) + \gamma_1\delta(s_i,1)) \delta(s_i,s_j) - \gamma_{01}(1 - \delta(s_i, s_j)) ]\delta(\sigma_i, \sigma_j)\]

In terms of communities this is

\[Q = \sum_c (e_c - \gamma_{01} 2 n_c(0) n_c(1) - \gamma_0 n^2_c(0) - \gamma_1 n^2_c(1))\]

where \(n_c(0)\) is the number of nodes in community \(c\) of class 0 (and similarly for 1) and \(e_c\) is the number of edges within community \(c\). We denote by \(n_c = n_c(0) + n_c(1)\) the total number of nodes in community \(c\).

We achieve this by creating three layers : (1) all nodes have node_size = 1 and all relevant links; (2) only nodes of class 0 have node_size = 1 and no links; (3) only nodes of class 1 have node_size = 1 and no links. If we add the first with resolution parameter \(\gamma_{01}\), and the others with resolution parameters \(\gamma_{01} - \gamma_0\) and \(\gamma_{01} - \gamma_1\), but the latter two with a layer weight of -1 while the first layer has layer weight 1, we obtain the following:

\[\begin{split}Q &= \sum_c (e_c - \gamma_{01} n_c^2) -\sum_c (- (\gamma_{01} - \gamma_0) n_c(0)^2) -\sum_c (- (\gamma_{01} - \gamma_1) n_c(1)^2) \\ &= \sum_c [e_c - \gamma_{01} 2 n_c(0) n_c(1) - \gamma_{01} n_c(0)^2 - \gamma_{01} n_c(1)^2) + ( \gamma_{01} - \gamma_0) n_c(0)^2 + ( \gamma_{01} - \gamma_1) n_c(1)^2 ] \\ &= \sum_c [e_c - \gamma_{01} 2 n_c(0) n_c(1) - \gamma_{0} n_c(0)^2 - \gamma_{1} n_c(1)^2]\end{split}\]

Although the derivation above is using \(n_c^2\), implicitly assuming a direct graph with self-loops, similar derivations can be made for undirected graphs using \(\binom{n_c}{2}\), but the notation is then somewhat more convoluted.

If we set node sizes equal to the degree, we get something similar to modularity, except that the resolution parameter should still be divided by \(2m\). In particular, in general (i.e. not specifically for bipartite graph) if node_sizes=G.degree() we then obtain

\[Q = \sum_{ij} A_{ij} - \gamma k_i k_j\]

In the case of bipartite graphs something similar is obtained, but then correctly adapted (as long as the resolution parameter is also appropriately rescaled).

Note

This function is not suited for directed graphs in the case of using the degree as node sizes.

SignificanceVertexPartition

class louvain.SignificanceVertexPartition(graph, initial_membership=None, node_sizes=None)

Bases: MutableVertexPartition

Implements Significance.

Notes

The quality function is

\[Q = \sum_c \binom{n_c}{2} D(p_c \parallel p)\]

where \(n_c\) is the number of nodes in community \(c\),

\[p_c = \frac{m_c}{\binom{n_c}{2}},\]

is the density of community \(c\),

\[p = \frac{m}{\binom{n}{2}}\]

is the overall density of the graph, and finally

\[D(x \parallel y) = x \ln \frac{x}{y} + (1 - x) \ln \frac{1 - x}{1 - y}\]

is the binary Kullback-Leibler divergence.

For directed graphs simply multiply the binomials by 2. The expected Significance in Erdos-Renyi graphs behaves roughly as \(\frac{1}{2} n \ln n\) for both directed and undirected graphs in this formulation.

Warning

This method is not suitable for weighted graphs.

References

1

Traag, V. A., Krings, G., & Van Dooren, P. (2013). Significant scales in community structure. Scientific Reports, 3, 2930. 10.1038/srep02930

Parameters

• graph (ig.Graph) – Graph to define the partition on.

• initial_membership (list of int) – Initial membership for the partition. If None then defaults to a singleton partition.

• node_sizes (list of int, or vertex attribute) – Sizes of nodes are necessary to know the size of communities in aggregate graphs. Usually this is set to 1 for all nodes, but in specific cases this could be changed.

SurpriseVertexPartition

class louvain.SurpriseVertexPartition(graph, initial_membership=None, weights=None, node_sizes=None)

Bases: MutableVertexPartition

Implements (asymptotic) Surprise.

Notes

The quality function is

\[Q = m D(q \parallel \langle q \rangle)\]

where \(m\) is the number of edges,

\[q = \frac{\sum_c m_c}{m},\]

is the fraction of internal edges,

\[\langle q \rangle = \frac{\sum_c \binom{n_c}{2}}{\binom{n}{2}}\]

is the expected fraction of internal edges, and finally

\[D(x \parallel y) = x \ln \frac{x}{y} + (1 - x) \ln \frac{1 - x}{1 - y}\]

is the binary Kullback-Leibler divergence.

For directed graphs we can multiplying the binomials by 2, and this leaves \(\langle q \rangle\) unchanged, so that we can simply use the same formulation. For weighted graphs we can simply count the total internal weight instead of the total number of edges for \(q\), while \(\langle q \rangle\) remains unchanged.

References

1

Traag, V. A., Aldecoa, R., & Delvenne, J.-C. (2015). Detecting communities using asymptotical surprise. Physical Review E, 92(2), 022816. 10.1103/PhysRevE.92.022816

Parameters

• graph (ig.Graph) – Graph to define the partition on.

• initial_membership (list of int) – Initial membership for the partition. If None then defaults to a singleton partition.

• weights (list of double, or edge attribute) – Weights of edges. Can be either an iterable or an edge attribute.

• node_sizes (list of int, or vertex attribute) – Sizes of nodes are necessary to know the size of communities in aggregate graphs. Usually this is set to 1 for all nodes, but in specific cases this could be changed.

• Index

• Search Page