def sample_runs(n_runs, single_steps, N, link_density, rewiring_prob, update_timescale, adj_mats):
n = 0
n_fails = 0
while n < n_runs:
print "Run:", n + 1
# Agent's parameters
MaxStocks = np.ones(N)
Stocks = np.random.rand(N) * MaxStocks
GrowthRates = np.ones(N)
Rationalities = np.ones(N)
Strategies = np.random.randint(2, size=N)
# Running the model
t0 = time()
m = Exploit(
adj_mats[n], Strategies, Stocks, MaxStocks, GrowthRates, Rationalities, rewiring_prob, update_timescale
)
if single_steps == -1:
fail = m.run() == 0
print "Runtime: {0} min".format((time() - t0) / 60.0)
if fail:
print "No conv., in run", n + 1
n_fails += 1
elif not fail:
print "<sus>: {}".format(m.get_strategies().mean())
n += 1
else:
m.run(steps=single_steps)
n += 1
print "Runtime: {0} min".format((time() - t0) / 60.0)
print "Done!"
def sample_runs(n_runs, single_steps, N, link_density, rewiring_prob,
update_timescale, adj_mats):
n = 0
n_fails = 0
while n < n_runs:
print "Run:", n + 1
# Agent's parameters
MaxStocks = np.ones(N)
Stocks = np.random.rand(N) * MaxStocks
GrowthRates = np.ones(N)
Rationalities = np.ones(N)
Strategies = np.random.randint(2, size=N)
# Running the model
t0 = time()
m = Exploit(adj_mats[n], Strategies, Stocks, MaxStocks, GrowthRates,
Rationalities, rewiring_prob, update_timescale)
if single_steps == -1:
fail = (m.run() == 0)
print "Runtime: {0} min".format((time() - t0) / 60.)
if fail:
print "No conv., in run", n + 1
n_fails += 1
elif not fail:
print "<sus>: {}".format(m.get_strategies().mean())
n += 1
else:
m.run(steps=single_steps)
n += 1
print "Runtime: {0} min".format((time() - t0) / 60.)
print "Done!"
def test_consensus():
"""
Test the consensus method (and that it is properly called in the __init__)
for extremely simple cases.
"""
A = np.zeros((3, 3), dtype=int) # Dummy empty adjaceny matrix
ones = np.ones(3) # Dummy variable for stock, groth_rates etc.
# Two unconnected network components with different strategies
A[0, 1] = A[1, 0] = 1
S = np.array([1, 1, 0])
m = ExploitCore(A, S, ones, ones, ones, ones, 0.5, 2.0)
assert m.get_consensus() == True
# Both strategies in the same component
S = np.array([1, 0, 1])
m = ExploitCore(A, S, ones, ones, ones, ones, 0.5, 2.0)
assert m.get_consensus() == False
# One component with one strategy
A[0, 2] = A[2, 0] = 1
S = np.array([1, 1, 1])
m = ExploitCore(A, S, ones, ones, ones, ones, 0.5, 2.0)
assert m.get_consensus() == True
def test_getters():
"""
Test if the assignement and return of the adjacency matrix works
"""
A = np.random.randint(2, size=(10, 10))
S = np.random.randint(2, size=10)
stocks = np.random.rand(10)
ones = np.ones(10)
# Dummy values
m = ExploitCore(A, S, stocks, ones, ones, ones, 0.5, 2.0)
assert (m.get_adjacency() == A).all()
assert (m.get_strategies() == S).all()
assert (m.get_stocks() == stocks).all()
assert m.get_time() == 0
def test_update():
"""
Test if one update process changes either the adjacency matrix or the
strategies.
"""
A = nx.adj_matrix(nx.erdos_renyi_graph(10, 0.3)).toarray()
S = np.random.randint(2, size=10)
stocks = np.ones(10)
ones = np.ones(10)
# If rewire (phi=1) A has to change
m = ExploitCore(A.copy(), S, stocks.copy(), ones, ones, ones, 1.0, 2.0)
m.run(steps=1)
assert np.sum(np.abs(m.get_adjacency() - A)) == 4
# Stocks decreased
assert (m.get_stocks() < stocks).all()
# If no rewirte (phi=0) S has to change
m = ExploitCore(A, S.copy(), stocks.copy(), ones, ones, ones, 0.0, 2.0)
m.run(steps=1)
assert np.sum(np.abs(m.get_strategies() - S)) == 1
# Stocks decreased
assert (m.get_stocks() < stocks).all()
def RUN_FUNC(tau, phi, sigma, mu, N, link_density, g, gamma, filename):
"""
Setting up the model with various parameters and determine what and how to
save the outcome. This is done with a pickeld dictionary including the
inital values, some parameters (for double checking) and the consensous
state and time - if the model reached consensus.
Parameters
----------
tau : float
the social update timescale
phi : float
the rewiring probability (between 0 and 1)
sigma : float
the standard deviation of the normal distribution for the capacities
mu : float
the mean of the normal distribution for the capacities
N : int
the number of agents/resources
link_density : float
the link density of the erdoes-renyi network
g : float
the (homogenous) growth rate of all resources
gamma : float
the (homogenous) rationality of all agents
filename : string
the filename where to pickle - passed from "experiment_handling"
This needs to be the last parameter of the RUN_FUNC
"""
MaxStocks = np.abs(np.random.normal(mu, sigma, size=N)) # normal
Stocks = np.random.rand(N) * MaxStocks # Random inital stocks
Strategies = np.random.randint(2, size=N) # Random inital strategies
GrowthRates = g * np.ones(N) # Homogenous growth rates g
Rationalities = gamma * np.ones(N) # Homogenous rationalities gamma
# Connected adjacnecy matrix
net = nx.erdos_renyi_graph(N, link_density)
while len(list(nx.connected_components(net))) > 1:
print "Network has isolated components. Try again!"
net = nx.erdos_renyi_graph(N, link_density)
A = nx.adj_matrix(net).toarray()
# Prepare storing inititals
res = {}
res["initials"] = pd.DataFrame({
"MaxStocks": MaxStocks,
"Stocks": Stocks,
"Strategies": Strategies
})
# Prpare storing parameters
res["parameters"] = pd.Series({
"N": N,
"link_density": link_density,
"growth_rate": g,
"rationality": gamma,
"mu_lognorm": mu,
"sigma_lognorm": sigma,
"rewiring_prob": phi,
"update_timescale": tau
})
# Running the model
m = Exploit(A, Strategies, Stocks, MaxStocks, GrowthRates, Rationalities,
phi, tau)
exit_status = m.run()
if exit_status == 1:
# Prepare storing consensus state
res["consensus"] = pd.DataFrame({
"Strategies": m.get_strategies(),
"Stocks": m.get_stocks()
})
res["consensus_time"] = m.get_time()
cPickle.dump(res, open(filename, "wb"))
return exit_status
def RUN_FUNC(tau, phi, sigma, C, N, link_density, g, gamma, filename):
"""
Setting up the model with various parameters and determine what and how to
save the outcome. This is done with a pickeld dictionary including the
inital values, some parameters (for double checking) and the consensous
state and time - if the model reached consensus.
Parameters
----------
tau : float
the social update timescale
phi : float
the rewiring probability (between 0 and 1)
sigma : float
the sigma parameter of the lognormal distribution for the capacities
C : float
C = mu + sigma^2/2 of the lognormal distribution. By that the
distributions of the resouces have equal mean
N : int
the number of agents/resources
link_density : float
the link density of the erdoes-renyi network
g : float
the (homogenous) growth rate of all resources
gamma : float
the (homogenous) rationality of all agents
filename : string
the filename where to pickle - passed from "experiment_handling"
This needs to be the last parameter of the RUN_FUNC
"""
mu = float(C) - sigma**2/2.
MaxStocks = np.random.lognormal(mu, sigma, size=N) # log-normal
Stocks = np.random.rand(N) * MaxStocks # Random inital stocks
Strategies = np.random.randint(2, size=N) # Random inital strategies
GrowthRates = g * np.ones(N) # Homogenous growth rates g
Rationalities = gamma * np.ones(N) # Homogenous rationalities gamma
# Connected adjacnecy matrix
net = nx.erdos_renyi_graph(N, link_density)
while len(list(nx.connected_components(net))) > 1:
print "Network has isolated components. Try again!"
net = nx.erdos_renyi_graph(N, link_density)
A = nx.adj_matrix(net).toarray()
# Prepare storing inititals
res = {}
res["initials"] = pd.DataFrame({"MaxStocks": MaxStocks,
"Stocks": Stocks,
"Strategies": Strategies})
# Prpare storing parameters
res["parameters"] = pd.Series({"N": N,
"link_density": link_density,
"growth_rate": g,
"rationality": gamma,
"mu_lognorm": mu,
"sigma_lognorm": sigma,
"rewiring_prob": phi,
"update_timescale": tau})
# Running the model
m = Exploit(A, Strategies, Stocks, MaxStocks, GrowthRates,
Rationalities, phi, tau)
exit_status = m.run()
if exit_status == 1:
# Prepare storing consensus state
res["consensus"] = pd.DataFrame({"Strategies": m.get_strategies(),
"Stocks": m.get_stocks()})
res["consensus_time"] = m.get_time()
cPickle.dump(res, open(filename, "wb"))
return exit_status
