nxsim

nxsim is a Python package for simulating agents in a network.

License: Apache Software License

Tags: networks, simulation, complex networks, discrete event simulation, forward simulation, temporal networks

nxsim

Nxsim is a Python package for simulating agents connected by any type of network using SimPy and Networkx in Python 3.4.

Install

pip3 install nxsim  # from PyPI
pip3 install git+git://github.com/kentwait/nxsim.git  # from GitHub

Quickstart

Nxsim provides a framework for doing forward-time simulations of events occurring in a network. It uses Networkx to create a network and SimPy 3 to create agents over each node in the network.

To create a simulation, nxsim requires a graph generated by Networkx and an “agent” class to populate each node of the network.

First, create a graph using Networkx.

import networkx as nx

number_of_nodes = 10
G = nx.complete_graph(number_of_nodes)

Then, subclass BaseNetworkAgent to create your own agent based on your needs.

from nxsim import BaseNetworkAgent

# Just like subclassing a process in SimPy
class MyAgent(BaseNetworkAgent):
    def __init__(self, environment=None, agent_id=0, state=()):  # Make sure to have these three keyword arguments
        super().__init__(environment=environment, agent_id=agent_id, state=state)
        # Add your own attributes here

    def run(self):
        # Add your behaviors here

Notice that “agents” in nxsim use the same concepts as “processes” in SimPy 3 except that their interactions can be limited by the graph in the simulation environment. For more information about SimPy, they have a great introduction posted on their website.

Here is a graph-based example:

import random
from nxsim import BaseNetworkAgent

class ZombieOutbreak(BaseNetworkAgent):
    def __init__(self, environment=None, agent_id=0, state=()):
        super().__init__(environment=environment, agent_id=agent_id, state=state)
        self.bite_prob = 0.05

    def run(self):
        while True:
            if self.state['id'] == 1:
                self.zombify()
                yield self.env.timeout(1)
            else:
                yield self.env.event()

    def zombify(self):
        normal_neighbors = self.get_neighboring_agents(state_id=0)
        for neighbor in normal_neighbors:
            if random.random() < self.bite_prob:
                neighbor.state['id'] = 1  # zombie
                print(self.env.now, self.id, neighbor.id, sep='\t')
                break

You can now set-up your simulation by creating a NetworkSimulation instance.

from nxsim import NetworkSimulation

# Initialize agent states. Let's assume everyone is normal.
# Add keys as as necessary, but "id" must always refer to that state category
init_states = [{'id': 0, } for _ in range(number_of_nodes)]

# Seed a zombie
init_states[5] = {'id': 1}
sim = NetworkSimulation(topology=G, states=init_states, agent_type=ZombieOutbreak,
                        max_time=30, dir_path='sim_01', num_trials=1, logging_interval=1.0)

And finally, start it up.

sim.run_simulation()

Running the simulation saves pickled dictionaries into the dir_path folder, in this case to “sim_01”. Now, let’s retrieve the history and states of the trial

trial = BaseLoggingAgent.open_trial_state_history(dir_path='sim_01', trial_id=0)

And plot the number of zombies per time interval using matplotlib:

from matplotlib import pyplot as plt
zombie_census = [sum([1 for node_id, state in g.items() if state['id'] == 1]) for t,g in trial.items()]
plt.plot(zombie_census)

And that’s it!