def testWeightedIndexRespectsDistribution(self):
# TODO: This is a change detector; find a better strategy than RNG seeding.
numpy.random.seed(0)
distribution = numpy.array([0.3, 0.0, 0.3, 0.3, 0.1])
expected = [2, 2, 2, 2, 2, 3, 3, 3, 3, 4]
self.assertItemsEqual(
expected, helpers.weighted_index(distribution, len(expected)))
def testWeightedIndexDefaultNumberResults(self):
distribution = numpy.array([0.0, 1.0, 0.0, 0.0])
# 1 until the end of time.
expected = numpy.array([1, 1, 1, 1])
numpy.testing.assert_equal(expected,
helpers.weighted_index(distribution))
def get_signal(self, meaning_index):
"""Returns a signal by sampling from the P matrix.
Arguments:
meaning_index: Index of the meaning for which to produce a signal.
Returns:
The index of a signal determined by the weights in the active matrix.
Raises:
IndexError if meaning_index >= number of signals or meaningy_index < 0.
"""
return helpers.weighted_index(self._active_matrix[meaning_index], 1)[0]
def get_meaning(self, signal_index):
"""Returns a meaning by sampling from the Q matrix.
Arguments:
signal_index: Index of the signal for which to produce a meaning.
Returns:
The index of a meaning determined by the weights in the passive matrix.
Raises:
IndexError if signal_index >= number of meanings or signal_index < 0.
"""
return helpers.weighted_index(self._passive_matrix[signal_index], 1)[0]
def reproduce(state):
payoff_vector = numpy.vectorize(
lambda individual: fitness.total_payoff(individual, state.
individuals))
payoffs = payoff_vector(state.individuals)
normalized = helpers.safe_divide(payoffs, numpy.sum(payoffs))
new_indices = helpers.weighted_index(normalized)
# TODO: Create a vfunc to create a freq dict from new_indices, then another
# to cache association matrices, obviating duplicate calculation. Then
# again, maybe each child SHOULD get a separate A matrix?
parents = state.individuals[new_indices]
return numpy.array([
model.Interlocutor.from_association_matrix(reveal(revelator, k))
for revelator in parents
])