def create_households(self, land, demand):
# calculate the fleet and household size table
fleet_array = np.array([freq for fleet, freq in self.fleet_freq])
size_array = np.array([freq for size, freq in self.size_freq])
# fill the joint fleet-hhsize table using proportional fitting
table = self._proportional_fit(fleet_array, size_array)
# calculate the number of households
hhnum = int(round(table.sum()))
# calculate the number of workers
# if the size of a household is one, this one person is a worker
# if the size of a household is larger than one, there a two workers
wknum = sum(((2 if size > 3 else size) * freq for size, freq in self.size_freq))
# calculate the number of students, all the other persons are students
stnum = sum(((0 if size < 3 else size - 2) * freq for size, freq in self.size_freq))
# assign random activity program to the households
program_freq = dict(((demand.programs[id_], freq) for id_, freq in self.prog_freq))
programs = self._get_assignments(program_freq, hhnum)
# assign random dwelling unit to the households
residences = self._get_assignments(land.get_capacities("home"), hhnum)
# assign random work place to the workers
offices = self._get_assignments(land.get_capacities("work"), wknum)
# assgin random school to the students
schools = self._get_assignments(land.get_capacities("school"), stnum)
# create iterator for activity programs
it_program = iter(programs)
# create iterators for all the locations
it_residence, it_office, it_school = iter(residences), iter(offices), iter(schools)
# create a household pool
for i, j in ndrange(*table.shape):
# create households with the same size and fleet
for _ in xrange(int(round(table[i, j]))):
self.add_household(self.size_freq[j][0], self.fleet_freq[i][0],
demand, it_program, it_residence, it_office, it_school)
def _proportional_fit(self, row_sum, col_sum, tolerance=0.01):
n_row, n_col = row_sum.size, col_sum.size
# convert to matrices
row_sum.shape = (n_row, 1)
col_sum.shape = (1, n_col)
# the row sum and column sum should be equal
assert row_sum.sum() == col_sum.sum(
), 'Row subsum and column subsum are not equal.'
# initialize a table
table = np.ones([n_row, n_col])
# this table is a upper triangular matrix
for i, j in ndrange(n_row, n_col):
if i > j:
table[i, j] = 0.0
row_err = float('+inf')
# check convergence criteria
while row_err > tolerance:
# row proportional fitting
table = row_sum * (table / table.sum(1).reshape(n_row, 1))
# column proportional fitting
table = col_sum * (table / table.sum(0).reshape(1, n_col))
# calculate the differences
row_diff = table.sum(1).reshape(n_row, 1) - row_sum
row_err = (row_diff * row_diff).sum()
return table
def create_households(self, land, demand):
# calculate the fleet and household size table
fleet_array = np.array([freq for fleet, freq in self.fleet_freq])
size_array = np.array([freq for size, freq in self.size_freq])
# fill the joint fleet-hhsize table using proportional fitting
table = self._proportional_fit(fleet_array, size_array)
# calculate the number of households
hhnum = int(round(table.sum()))
# calculate the number of workers
# if the size of a household is one, this one person is a worker
# if the size of a household is larger than one, there a two workers
wknum = sum(((2 if size > 3 else size) * freq
for size, freq in self.size_freq))
# calculate the number of students, all the other persons are students
stnum = sum(((0 if size < 3 else size - 2) * freq
for size, freq in self.size_freq))
# assign random activity program to the households
program_freq = dict(
((demand.programs[id_], freq) for id_, freq in self.prog_freq))
programs = self._get_assignments(program_freq, hhnum)
# assign random dwelling unit to the households
residences = self._get_assignments(land.get_capacities("home"), hhnum)
# assign random work place to the workers
offices = self._get_assignments(land.get_capacities("work"), wknum)
# assgin random school to the students
schools = self._get_assignments(land.get_capacities("school"), stnum)
# create iterator for activity programs
it_program = iter(programs)
# create iterators for all the locations
it_residence, it_office, it_school = iter(residences), iter(
offices), iter(schools)
# create a household pool
for i, j in ndrange(*table.shape):
# create households with the same size and fleet
for _ in xrange(int(round(table[i, j]))):
self.add_household(self.size_freq[j][0], self.fleet_freq[i][0],
demand, it_program, it_residence, it_office,
it_school)
def _proportional_fit(self, row_sum, col_sum, tolerance=0.01):
n_row, n_col = row_sum.size, col_sum.size
# convert to matrices
row_sum.shape = (n_row, 1)
col_sum.shape = (1, n_col)
# the row sum and column sum should be equal
assert row_sum.sum() == col_sum.sum(), 'Row subsum and column subsum are not equal.'
# initialize a table
table = np.ones([n_row, n_col])
# this table is a upper triangular matrix
for i, j in ndrange(n_row, n_col):
if i > j:
table[i,j] = 0.0
row_err = float('+inf')
# check convergence criteria
while row_err > tolerance:
# row proportional fitting
table = row_sum * (table / table.sum(1).reshape(n_row, 1))
# column proportional fitting
table = col_sum * (table / table.sum(0).reshape(1, n_col))
# calculate the differences
row_diff = table.sum(1).reshape(n_row, 1) - row_sum
row_err = (row_diff*row_diff).sum()
return table