"""Node redundancy for bipartite graphs."""
from itertools import combinations

from networkx import NetworkXError

__all__ = ["node_redundancy"]


def node_redundancy(G, nodes=None):
    r"""Computes the node redundancy coefficients for the nodes in the bipartite
    graph `G`.

    The redundancy coefficient of a node `v` is the fraction of pairs of
    neighbors of `v` that are both linked to other nodes. In a one-mode
    projection these nodes would be linked together even if `v` were
    not there.

    More formally, for any vertex `v`, the *redundancy coefficient of `v`* is
    defined by

    .. math::

        rc(v) = \frac{|\{\{u, w\} \subseteq N(v),
        \: \exists v' \neq  v,\: (v',u) \in E\:
        \mathrm{and}\: (v',w) \in E\}|}{ \frac{|N(v)|(|N(v)|-1)}{2}},

    where `N(v)` is the set of neighbors of `v` in `G`.

    Parameters
    ----------
    G : graph
        A bipartite graph

    nodes : list or iterable (optional)
        Compute redundancy for these nodes. The default is all nodes in G.

    Returns
    -------
    redundancy : dictionary
        A dictionary keyed by node with the node redundancy value.

    Examples
    --------
    Compute the redundancy coefficient of each node in a graph::

        >>> from networkx.algorithms import bipartite
        >>> G = nx.cycle_graph(4)
        >>> rc = bipartite.node_redundancy(G)
        >>> rc[0]
        1.0

    Compute the average redundancy for the graph::

        >>> from networkx.algorithms import bipartite
        >>> G = nx.cycle_graph(4)
        >>> rc = bipartite.node_redundancy(G)
        >>> sum(rc.values()) / len(G)
        1.0

    Compute the average redundancy for a set of nodes::

        >>> from networkx.algorithms import bipartite
        >>> G = nx.cycle_graph(4)
        >>> rc = bipartite.node_redundancy(G)
        >>> nodes = [0, 2]
        >>> sum(rc[n] for n in nodes) / len(nodes)
        1.0

    Raises
    ------
    NetworkXError
        If any of the nodes in the graph (or in `nodes`, if specified) has
        (out-)degree less than two (which would result in division by zero,
        according to the definition of the redundancy coefficient).

    References
    ----------
    .. [1] Latapy, Matthieu, Clémence Magnien, and Nathalie Del Vecchio (2008).
       Basic notions for the analysis of large two-mode networks.
       Social Networks 30(1), 31--48.

    """
    if nodes is None:
        nodes = G
    if any(len(G[v]) < 2 for v in nodes):
        raise NetworkXError(
            "Cannot compute redundancy coefficient for a node"
            " that has fewer than two neighbors."
        )
    # TODO This can be trivially parallelized.
    return {v: _node_redundancy(G, v) for v in nodes}


def _node_redundancy(G, v):
    """Returns the redundancy of the node `v` in the bipartite graph `G`.

    If `G` is a graph with `n` nodes, the redundancy of a node is the ratio
    of the "overlap" of `v` to the maximum possible overlap of `v`
    according to its degree. The overlap of `v` is the number of pairs of
    neighbors that have mutual neighbors themselves, other than `v`.

    `v` must have at least two neighbors in `G`.

    """
    n = len(G[v])
    # TODO On Python 3, we could just use `G[u].keys() & G[w].keys()` instead
    # of instantiating the entire sets.
    overlap = sum(
        1 for (u, w) in combinations(G[v], 2) if (set(G[u]) & set(G[w])) - {v}
    )
    return (2 * overlap) / (n * (n - 1))