def test_simulate_discrite_distribution(self):
self.assertEqual(utils.simulate_discrete_distribution([1.0, 0.0]), 0, "Simulating singleton distribution fails")
self.assertEqual(utils.simulate_discrete_distribution([0.0, 1.0]), 1, "Simulating singleton distribution fails")
coin = [0.5, 0.5]
coin_tossing = np.array([utils.simulate_discrete_distribution(coin) for _ in xrange(0,500000)])
sample = np.mean(coin_tossing)
self.assertAlmostEqual(0.5, sample, delta=0.01, msg="Coin tossing test fails " + str(sample) +" is not 0.5" )
def _detached_step(population_array,
population_size,
payoff_function,
intensity_of_selection,
number_of_strategies,
mutation_kernel,
fitness_mapping='exp',
mutation_step=True):
"""
One generation step. This is a copy of the previous but it is detached so that it can be called more efficiently
Parameters:
----------
population_array =
mutation_step =
"""
current_distribution = (1.0 / population_size) * np.array(population_array,
dtype=float)
# COMPUTE PAYOFF
payoff = payoff_function(population_array)
# compute fitness distribution
fitness_sum = 0.0
fitness = np.empty_like(payoff)
if fitness_mapping == 'lin':
for i in range(0, number_of_strategies):
val = current_distribution[i] * (
1.0 - intensity_of_selection +
intensity_of_selection * payoff[i])
fitness[i] = val
fitness_sum += val
fitness /= fitness_sum
else:
for i in range(0, number_of_strategies):
val = current_distribution[i] * (the_number_e**(
intensity_of_selection * payoff[i]))
fitness[i] = val
fitness_sum += val
fitness /= fitness_sum
# choose one random guy in proportion to fitness
chosen_one = utils.simulate_discrete_distribution(fitness)
# mutate this guy
if mutation_step:
chosen_one = utils.simulate_discrete_distribution(
mutation_kernel[chosen_one])
# a random guy dies
dies = utils.simulate_discrete_distribution(current_distribution)
population_array[dies] -= 1
# we add a copy of the fit guy
population_array[chosen_one] += 1
# the payoff is returned but almost never used.
return StepResult(population_array, payoff)
def test_simulate_discrite_distribution(self):
self.assertEqual(utils.simulate_discrete_distribution([1.0, 0.0]), 0,
"Simulating singleton distribution fails")
self.assertEqual(utils.simulate_discrete_distribution([0.0, 1.0]), 1,
"Simulating singleton distribution fails")
coin = [0.5, 0.5]
coin_tossing = np.array([
utils.simulate_discrete_distribution(coin)
for _ in range(0, 500000)
])
sample = np.mean(coin_tossing)
self.assertAlmostEqual(0.5,
sample,
delta=0.01,
msg="Coin tossing test fails " + str(sample) +
" is not 0.5")
def _detached_step(population_array, population_size, payoff_function, intensity_of_selection, number_of_strategies,
mutation_kernel, fitness_mapping='exp', mutation_step=True):
"""
One generation step. This is a copy of the previous but it is detached so that it can be called more efficiently
Parameters:
----------
population_array =
mutation_step =
"""
current_distribution = (1.0 / population_size) * np.array(population_array, dtype=float)
# COMPUTE PAYOFF
payoff = payoff_function(population_array)
# compute fitness distribution
fitness_sum = 0.0
fitness = np.empty_like(payoff)
if fitness_mapping == 'lin':
for i in range(0, number_of_strategies):
val = current_distribution[i] * (1.0 - intensity_of_selection + intensity_of_selection * payoff[i])
fitness[i] = val
fitness_sum += val
fitness /= fitness_sum
else:
for i in range(0, number_of_strategies):
val = current_distribution[i] * (the_number_e ** (intensity_of_selection * payoff[i]))
fitness[i] = val
fitness_sum += val
fitness /= fitness_sum
# choose one random guy in proportion to fitness
chosen_one = utils.simulate_discrete_distribution(fitness)
# mutate this guy
if mutation_step:
chosen_one = utils.simulate_discrete_distribution(mutation_kernel[chosen_one])
# a random guy dies
dies = utils.simulate_discrete_distribution(current_distribution)
population_array[dies] -= 1
# we add a copy of the fit guy
population_array[chosen_one] += 1
# the payoff is returned but almost never used.
return StepResult(population_array, payoff)
def test_simulate_discrete_distribution_longer(self):
dist = [0.75, 0.2, 1.0-0.2-0.75]
zero = 0
one =0
two =0
for _ in xrange(0,5000000):
sample = utils.simulate_discrete_distribution(dist)
if sample ==0:
zero+=1
if sample ==1:
one+=1
if sample ==2:
two+=1
result = [zero, one, two]
result /= np.sum(result, dtype=float)
np.testing.assert_allclose(dist, result, rtol=0.001, atol=0.0001, err_msg="Bad simulate discrete dist", verbose=True)
def test_simulate_discrete_distribution_longer(self):
dist = [0.75, 0.2, 1.0 - 0.2 - 0.75]
zero = 0
one = 0
two = 0
for _ in range(0, 5000000):
sample = utils.simulate_discrete_distribution(dist)
if sample == 0:
zero += 1
if sample == 1:
one += 1
if sample == 2:
two += 1
result = [zero, one, two]
result /= np.sum(result, dtype=float)
np.testing.assert_allclose(dist,
result,
rtol=0.001,
atol=0.0001,
err_msg="Bad simulate discrete dist",
verbose=True)