def run_social_circles_trial(
args: Tuple[Dict[networkgen.Agent, int], Tuple[int, int], Disease, Any]
) -> Optional[Tuple[float, float, float]]:
agent_to_quantity, grid_dims, disease, rand = args
sim_len = 200
sims_per_trial = 150
target_communities = 25
sc_results = networkgen.make_social_circles_network(agent_to_quantity,
grid_dims,
rand=rand)
if sc_results is None:
return None
net, _, = sc_results
if nx.number_connected_components(net.G) > target_communities:
return None
to_flicker = partitioning.fluidc_partition(net.G, target_communities)
proportion_flickering = len(to_flicker) / net.E
social_good_score = rate_social_good(net)
network_behavior = StaticFlickerBehavior(net.M, to_flicker, (True, False),
"Probs don't change this")
avg_sus = np.mean([
np.sum(
simulate(net.M, make_starting_sir(net.N, 1), disease,
network_behavior, sim_len, None, rand)[-1][0] > 0)
for _ in range(sims_per_trial)
]) / net.N
return proportion_flickering, avg_sus, social_good_score
def run_inf_prob_vs_perc_sus(name: str, diseases: Sequence[Disease],
new_network: Callable[[], Network],
flicker_config: RandomFlickerConfig,
num_trials: int, rng):
"""
Save the plot and csv of infection probability vs percent susceptible.
"""
results: Dict[float, List[float]] = defaultdict(lambda: [])
print(f'Running {name}')
pbar = tqdm(total=num_trials * len(diseases))
for disease, _ in it.product(diseases, range(num_trials)):
net = new_network()
flicker = flicker_config.make_behavior(net.M, net.intercommunity_edges)
sir0 = make_starting_sir(net.N, 1, rng)
perc_sus = np.sum(
simulate(net.M, sir0, disease, flicker, 100, None, rng)[-1][0] > 0
) / net.N
results[disease.trans_prob].append(perc_sus)
pbar.update()
x_coords = tuple(results.keys())
collected_data = np.array([list(values) for values in results.values()])
quartiles = np.quantile(collected_data, (.25, .75),
axis=1,
interpolation='midpoint')
y_coords = np.mean(collected_data, axis=1)
plt.figure()
plt.title(name)
plt.xlim(0, 1)
plt.ylim(0, 1)
plt.xlabel('Infection Probability')
plt.ylabel('Survival Percentage')
plt.plot(x_coords, y_coords)
plt.fill_between(x_coords, quartiles[0], quartiles[1], alpha=.4)
plt.savefig(f'results/{name}.png', format='png', dpi=200)
def run_agent_generated_trial(args: Tuple[Disease, AgentBehavior, int, Any])\
-> Optional[Tuple[float, float, float]]:
"""
args: (disease to use in the simulation,
the behavior agents have when generating the network,
the number of agents in the network,
an instance of np.random.default_rng)
"""
disease, agent_behavior, N, rand = args
sim_len = 200
sims_per_trial = 150
net = make_agent_generated_network(N, agent_behavior)
if net is None:
return None
to_flicker = partitioning.fluidc_partition(net.G, 50)
proportion_flickering = len(to_flicker) / net.E
social_good = rate_social_good(net)
network_behavior = StaticFlickerBehavior(net.M, to_flicker, (True, False),
"Probs don't change this")
avg_sus = np.mean([
np.sum(
simulate(net.M, make_starting_sir(net.N, 1), disease,
network_behavior, sim_len, None, rand)[-1][0] > 0)
for _ in range(sims_per_trial)
]) / net.N
return proportion_flickering, avg_sus, social_good
def run_connected_community_trial(args: Tuple[ConnectedCommunityConfiguration, Disease, Any])\
-> Optional[Tuple[float, float, float]]:
"""
args: (ConnectedCommunityConfiguration to use,
disease to use,
default_rng instance)
return: (proportion of edges that flicker, the average number of remaining susceptible agents)
or None on failure.
"""
sim_len = 200
sims_per_trial = 150
configuration, disease, rand = args
inner_degrees = configuration.make_inner_degrees()
outer_degrees = configuration.make_outer_degrees()
net = networkgen.make_connected_community_network(inner_degrees, outer_degrees, rand)
# If a network couldn't be successfully generated, return None to signal the failure
if net is None:
return None
to_flicker = {(u, v) for u, v in net.edges if net.communities[u] != net.communities[v]}
proportion_flickering = len(to_flicker) / net.E
social_good = rate_social_good(net)
behavior = StaticFlickerBehavior(net.M, to_flicker, (True, False), "Probs don't change this")
avg_sus = np.mean([np.sum(simulate(net.M, make_starting_sir(net.N, 1),
disease, behavior, sim_len, None, rand)[-1][0] > 0)
for _ in range(sims_per_trial)]) / net.N
return proportion_flickering, avg_sus, social_good
def experiment(make_network, seed):
"""
return: list of (high_reach_survival_rate, high_reach_sim_time,
low_reach_survival_rate, low_reach_sim_time)
"""
rng = np.random.default_rng(seed)
hr_net = make_network(high_reach)
lr_net = make_network(low_reach)
sir0 = make_starting_sir(hr_net.N, 1, rng)
hr_sirs = simulate(hr_net.M, sir0, disease,
make_behavior(high_reach, make_network, rng),
sim_len_cap, rng, None)
lr_sirs = simulate(lr_net.M, sir0, disease,
make_behavior(low_reach, make_network, rng),
sim_len_cap, rng, None)
return (calc_survival_rate(hr_sirs), len(hr_sirs),
calc_survival_rate(lr_sirs), len(lr_sirs))
def cc_pressure_vs_none_entry_point():
rng = np.random.default_rng(0xbeefee)
num_trials = 1000
disease = Disease(4, .4)
inner_bounds = 1, 15
outer_bounds = 1, 5
community_size = 20
n_communities = 25
make_ccn = MakeConnectedCommunity(community_size, inner_bounds,
n_communities, outer_bounds, rng)
pressure_configurations = [
SimplePressureConfig(radius, prob, rng)
for radius, prob in it.product((1, 2, 3), (.25, .5, .75))
]
# pressure_experiment(make_ccn, pressure_configurations, disease, num_trials, rng)
net = make_ccn()
simulate(net.M, make_starting_sir(net.N, 1, rng), disease,
pressure_configurations[1].make_behavior(net), 100,
nx.spring_layout(net.G))
def __call__(self, encoding: np.ndarray) -> float:
net = self._enc_to_G(encoding)
M = net.M
to_flicker = fluidc_partition(net.G, net.N//20)
flicker_behavior = StaticFlickerBehavior(M, to_flicker, (True, False), '')
avg_sus = np.mean([np.sum(simulate(M, make_starting_sir(len(M), 1),
self._disease, flicker_behavior, self._sim_len,
None, self._rand)[-1][0] > 0)
for _ in range(self._n_sims)]) / len(M)
cost = 2-avg_sus-rate_social_good(net, self._decay_func)
return cost
def main():
n_trials = 100
max_steps = 100
rng = np.random.default_rng(0)
disease = Disease(4, .2)
names = ('elitist-500', 'cavemen-50-10', 'spatial-network', 'cgg-500',
'watts-strogatz-500-4-.1')
paths = fio.network_names_to_paths(names)
behavior_configs = (RandomFlickerConfig(.5, 'Random .5', rng),
StaticFlickerConfig((True, False), 'Static .5'))
for net_name, path in zip(names, paths):
net = fio.read_network(path)
to_flicker = tuple((u, v) for u, v in net.edges
if net.communities[u] != net.communities[v])
proportion_flickering = len(to_flicker) / net.E
social_good = rate_social_good(net)
trial_to_pf = tuple(proportion_flickering for _ in range(n_trials))
trial_to_sg = tuple(social_good for _ in range(n_trials))
print(f'Running simulations for {net_name}.')
for config in behavior_configs:
behavior = config.make_behavior(net.M, to_flicker)
sim_results = [
get_final_stats(
simulate(net.M,
make_starting_sir(net.N, 1, rng),
disease,
behavior,
max_steps,
None,
rng=rng)) for _ in tqdm(range(n_trials))
]
results = BasicExperimentResult(f'{net_name} {config.name}',
sim_results, trial_to_pf,
trial_to_sg)
results.save_csv(RESULTS_DIR)
results.save_box_plots(RESULTS_DIR)
results.save_perc_sus_vs_social_good(RESULTS_DIR)
def run_experiments(args: Tuple[str, int, int, Disease, Sequence[FlickerConfig], str])\
-> Optional[FlickerComparisonResult]:
"""
Run a batch of experiments and return a tuple containing the network's name,
number of flickering edges, and a mapping of behavior name to the final
amount of susceptible nodes. Return None on failure.
args: (path to the network,
number of sims to run for each behavior,
simulation length,
disease,
a sequence of configs for the flickers to use,
the name of the baseline flicker to compare the other results to)
"""
network_path, num_sims, sim_len, disease, flicker_configs, baseline_flicker_name = args
net = fio.read_network(network_path)
intercommunity_edges = {(u, v)
for u, v in net.edges
if net.communities[u] != net.communities[v]}
behavior_to_results: Dict[str, Sequence[float]] = {}
for config in flicker_configs:
behavior = config.make_behavior(net.M, intercommunity_edges)
# The tuple comprehension is pretty arcane, so here is an explanation.
# Each entry is the sum of the number of entries in the final SIR where
# the days in S are greater than 0. That is to say, the number of
# susceptible agents at the end of the simulation.
perc_sus = tuple(
np.sum(
simulate(net.M, make_starting_sir(net.N, 1), disease, behavior,
sim_len, None)[-1][0] > 0) / net.N
for _ in range(num_sims))
behavior_to_results[behavior.name] = perc_sus
return FlickerComparisonResult(fio.get_network_name(network_path),
num_sims, sim_len,
len(intercommunity_edges) / net.E,
behavior_to_results, baseline_flicker_name)
def pressure_test_entry_point():
rng = np.random.default_rng()
net = fio.read_network('networks/cavemen-50-10.txt')
simulate(net.M, make_starting_sir(net.N, (0, ), rng), Disease(4, 0.3),
SimplePressureBehavior(net, 1), 200, None, rng)