Informations about types

As there are many ways to represent ecological networks, and the correct way to measure a given property varies in function of the representation, EcologicalNetwork has a series of built-in types.

Type hierarchy

The EcologicalNetwork package has six main types that are meant to be used by the user: BipartiteNetwork, BipartiteProbaNetwork, BipartiteQuantiNetwork, UnipartiteNetwork, UnipartiteProbaNetwork and UnipartiteQuantiNetwork. All types with Proba in their names are meant to represent probabilistic networks, and are also part of the union type ProbabilisticNetwork. All types without Proba in their name are part of the union type DeterministicNetwork, and represent networks in which interactions are either present or absent. All types with Quanti in their names are part of the QuantitativeNetwork groups, and represent networks with weighted interactions. All types starting with Bipartite are also part of the abstract type Bipartite, and types with Unipartite in their names are part of the abstract type Unipartite. Finally, both Unipartite and Bipartite are part of the abstract type EcoNetwork.

Although this may seem convoluted, this is important to understand: when writing functions, you can restrict them to any type of network you want by using the right type in their declaration. You can also check properties of a network just by looking at its type. For example, one can check whether a network N is bipartite with:

typeof(N) <: Bipartite

Networks are represented as two-dimensional matrices. All types are simply wrappers around an adjacency matrix, stored as the A property of the object. To look at the adjacency matrix of a network N, one therefore uses N.A. These matrices must be read as: the existence/probability of an interaction from the species of the i-th row to the species in the j-th column. This implies that the networks are, by default, directed.

Note that the type of a network will determine what methods can be applied to it. For example, all measures of variance are only making sense for probabilistic networks.

Data types

Interactions in deterministic networks are represented as boolean (true/false) values. This is memory efficient, so large networks can be represented (one interactions represented as a boolean uses 8 times fewer memory than the same interaction represented as an integer). This being said, all networks of the DeterministicNetwork type can be read from matrices of integers, as long as these matrices only contain 0 and 1.

In probabilistic networks, interactions are stored as floating point (Float64) numbers. These values have to be between 0.0 and 1.0, as they represent probabilities. In quantitative networks, interactions are stored as any type of number.

Networks of the Unipartite class must have the same number of rows and columns. The species in the rows and columns are the same. Networks of the Bipartite class are expected to have different numbers of rows and columns, as the species in rows and columns are different species. It is possible to convert a network from Bipartite to Unipartite using the make_unipartite function:

~~~@example using EcologicalNetwork B = BipartiteNetwork(rand(Bool, (3, 5))) U = make_unipartite(B) richness(U) == richness(B)

The documentation for `make_unipartite` gives additional explanations about how
the conversion is done. In the overwhelming majority of cases, applying any
measure to a bipartite network, and to the same network made unipartite, should
give the same results (connectance is one notable example).

## Type reference

~~~@autodocs
Modules = [EcologicalNetwork]
Order = [:type]
Pages = ["types.jl"]