README.rst

thresholdmodel

Simulate a continuous-time threshold model on static networks using Gillespie's stochastic simulation algorithm (SSA). The networks can be directed and/or weighted.

In contrast to the original discrete-time model, nodes whose aggregated inputs exceed their respective thresholds will not flip after the "next time step" because there are no time steps. Instead, a node whose threshold has been exceeded will enter an alert state from which it will enter the activated state with rate $gamma = 1$.

Install

pip install thresholdmodel

Example

Simulate on an ER random graph.

import numpy as np
import networkx as nx
import matplotlib.pyplot as plt

from thresholdmodel import ThreshModel

N = 1000
k = 10

thresholds = 0.1
initially_activated = np.arange(100)

G = nx.fast_gnp_random_graph(N, k/(N-1.0))

Thresh = ThreshModel(G,initially_activated,thresholds)
t, cascade_size = Thresh.simulate()

plt.plot(t,cascade_size)
plt.show()

API

Simulate

Given a networkx-Graph object G (can be a networkx.DiGraph, too), and values for initially_activated and thresholds, simulate like this

Thresh = ThreshModel(G,initially_activated,thresholds)
t, a = Thresh.simulate()

t is a numpy.ndarray containing the times at which node activations happened. a is a numpy.ndarray containing the relative cascade size at the corresponding time in t. Note that the whole process is modeled as a Poisson process such that the time t will be given in units of the node activation rate gamma = 1.0. If you want to simulate for another node activation rate, simply rescale time as t /= gamma.

When the simulation is started with the save_activated_nodes=True flag, a list of activated nodes per time leap is saved in ThreshModel.activated_nodes.

t, a = Thresh.simulate(save_activated_nodes=True)
print(Thresh.activated_nodes)

You can repeat a simulation with the same initial conditions by simply calling Thresh.simulate() again, all the necessary things will be reset automatically.

Set initially activated nodes

Set nodes 3, 5, and 8 to be activated initially.

initially_activated = [3, 5, 8] # this could also be a numpy array

Choose 20% of all nodes randomly to be activated initially. When the simulation is restarted, the same nodes will be chosen as initial conditions.

initially_activated = 0.2

Choose 35 randomly selected nodes to be activated initially. When the simulation is restarted, the same nodes will be chosen as initial conditions.

initially_activated = 35

Set thresholds

Activation thresholds can be set for all nodes

thresholds = np.random.rand(G.number_of_nodes())

Note that thresholds need to lie in the domain [0,1].

You can also set a universal threshold

thresholds = 0.1

Here, 10% of a node's neighbors need to be activated in order for the node to become active, too.

Directed networks

A node will become active if the sufficient number of nodes pointing towards the node are active. This means that a node's in-degree will be the important measure to determine wether this particular node will become active.

Weighted networks

If you want to simulate on a weighted network, provide the weight keyword

Thresh = ThreshModel(G,initially_activated,thresholds,weight='weight')

Similar to the networkx-documentation: weight (string, optional (default=None)) - The attribute name to obtain the edge weights. E.g.: G.edges[0,1]['weight'].

A focal node will become active when the cumulative edge weights of all activated nodes pointing towards the focal node will reach > threshold*in_degree.

Docstring

This is the model's docstring.

>>> help(ThreshModel)
Help on class ThreshModel in module thresholdmodel.model:

class ThreshModel(builtins.object)
 |  ThreshModel(G, initially_activated, thresholds, weight=None)
 |
 |  A simple simulation class that runs
 |  a threshold-model activation process
 |  on a static network (potentially weighted and directed)
 |  in continuous time using Gillespie's
 |  stochastic simulation algorithm.
 |
 |  The temporal dimension is fixed by assuming
 |  that every node whose activation threshold
 |  has been exceeded by neighboring inputs
 |  is activated with constant and uniform
 |  rate :math:`\gamma = 1`.
 |
 |  Parameters
 |  ==========
 |  G : networkx.Graph, networkx.DiGraph
 |      The network on which to simulate.
 |      Nodes must be integers in the range
 |      of ``[0, N-1]``.
 |  initially_activated: float, int, or list of ints
 |      Can be either of three things:
 |
 |      1. float of value ``0 < initially_activated < 1``.
 |         In this case, ``initially_activated`` is
 |         interpreted to represent a fraction of nodes
 |         that will be randomly selected from the
 |         set of nodes and set to be activated.
 |      2. int of value ``1 <= initially_activated < N-1``.
 |         In this case, ``initially_activated`` nodes
 |         will be randomly sampled from the node set
 |         and set to be activated.
 |      3. list of ints. In this case, ``initially_activated``
 |         is interpreted to contain indices of nodes
 |         that will be activated initially.
 |  thresholds: float or iterable of floats
 |      Can be either of two things:
 |
 |      1. float of value ``0 < thresholds <= 1``.
 |         In this case, every node will have the same
 |         activation threshold.
 |      2. iterable of values ``0 < thresholds <=1``.
 |         In this case, the function expectes a list,
 |         tuple, or array with length equal to the
 |         number of nodes. Each entry `m` of this list
 |         will be interpreted to be node `m`'s activation
 |         threshold.
 |  weight: str, default = None
 |      A string that represents the weight keyword of a link.
 |      If `None`, the network is assumed to be unweighted.
 |
 |  Example
 |  =======
 |
 |  >>> G = nx.fast_gnp_random_graph(1000,20/(1000-1))
 |  >>> model = TreshModel(G, 100, 0.1)
 |  >>> t, cascade_size = model.simulate()
 |
 |  Attributes
 |  ==========
 |  G : nx.Graph or nx.DiGraph
 |      The network on which to simulate.
 |      Nodes must be integers in the range
 |      of ``[0, N-1]``.
 |  N : int
 |      The number of nodes in the network
 |  weight: str
 |      A string that represents the weight keyword of a link.
 |      If `None`, the network is assumed to be unweighted.
 |  in_deg : numpy.ndarray
 |      Contains the in-degree of every node.
 |  thresholds: numpy.ndarray
 |      Each entry `m` of this array
 |      represents node `m`'s activation
 |      threshold.
 |  initially_activated: numpy.ndarray
 |      Each entry of this array contains
 |      a node that will be activated initially.
 |  time: numpy.ndarray
 |      Contains every time point at which a node was
 |      activates (after ``simulation()`` was called).
 |      The temporal dimension is given by assuming
 |      that every node whose activation threshold
 |      has been exceeded by activation inputs
 |      is activated with constant and uniform
 |      rate :math:`\gamma = 1`.
 |  cascade_size: numpy.ndarray
 |      The relative size of the activation cascade
 |      at the corrsponding time value in ``time``
 |      (relative to the size of the node set).
 |      Only available after ``simulation()`` was called.
 |  activated_nodes: list
 |      A list of lists.
 |      Each entry contains a list of integers representing
 |      the nodes that have been activated
 |      at the corrsponding time value in ``time``.
 |      Each list entry will contain only a single node
 |      for every other time than the initial time.
 |      Only available after ``simulation()`` was called.
 |
 |  Methods defined here:
 |
 |  __init__(self, G, initially_activated, thresholds, weight=None)
 |      Initialize self.  See help(type(self)) for accurate signature.
 |
 |  reset(self)
 |      Reset the simulation.
 |
 |  set_initially_activated(self, initially_activated)
 |      Set the process's initial activation state.
 |
 |      Parameters
 |      ==========
 |      initially_activated: float, int, or list of ints
 |          Can be either of three things:
 |
 |          1. float of value ``0 < initially_activated < 1``.
 |             In this case, ``initially_activated`` is
 |             interpreted to represent a fraction of nodes
 |             that will be randomly selected from the
 |             set of nodes and set to be activated.
 |          2. int of value ``1 <= initially_activated < N-1``.
 |             In this case, ``initially_activated`` nodes
 |             will be randomly sampled from the node set
 |             and set to be activated.
 |          3. list of ints. In this case, ``initially_activated``
 |             is interpreted to contain indices of nodes
 |             that will be activated initially.
 |
 |  set_thresholds(self, thresholds)
 |      Set node activation thresholds.
 |
 |      Parameters
 |      ==========
 |      thresholds: float or iterable of floats
 |          Can be either of two things:
 |
 |          1. float of value ``0 < thresholds <= 1``.
 |             In this case, every node will have the same
 |             activation threshold.
 |          2. iterable of values ``0 < thresholds <=1``.
 |             In this case, the function expectes a list,
 |             tuple, or array with length equal to the
 |             number of nodes. Each entry `m` of this list
 |             will be interpreted to be node `m`'s activation
 |             threshold.
 |
 |  simulate(self, save_activated_nodes=False)
 |      Simulate until all nodes that can be activated
 |      have been activated.
 |
 |      Parameters
 |      ==========
 |      save_activated_nodes: bool, default = False
 |          If ``True``, write a list of activated nodes
 |          to the class attribute ``activated_nodes``
 |          every time an activation event happens.
 |          Such a list will contain only a single node
 |          for every other time than the initial time.
 |
 |      Returns
 |      =======
 |      time : numpy.ndarray
 |          Time points at which nodes were activated.
 |      cascade_size : numpy.ndarray
 |          The relative size of the activation cascade
 |          at the corrsponding time value in ``time``
 |          (relative to the size of the node set).