def run_cras(row_totals: numpy.matrix,
column_totals: numpy.matrix,
conditions: dict,
maximum_iterations: int=10000,
tol: float=0.00001) -> numpy.matrix:
"""
:param row_totals:
:param column_totals:
:param conditions: a condition is a group of cells and a total, dict(tuple(tuples): float),
a cell is (row_index, col_index)
:param maximum_iterations: big number to stop this running forever
:param tol: how close is close enough?
:return:
"""
# TODO remove this hack [max(thing.shape)]
a = numpy.matrix(numpy.ones(shape=(max(row_totals.shape), max(column_totals.shape))))
e = numpy.matrix(numpy.ones(shape=(len(row_totals), 1)))
for _ in range(maximum_iterations):
r_hat = row_scaling(a, row_totals, e)
a = matrix_multiply(r_hat, a)
s_hat = column_scaling(a, column_totals, e)
a = matrix_multiply(a, s_hat)
a = apply_conditions(a, conditions)
if is_matrix_close_to_i(r_hat, tol) and is_matrix_close_to_i(s_hat, tol):
return a
return a
def run_ras(row_totals: numpy.matrix,
column_totals: numpy.matrix,
maximum_iterations: int=10000,
a: numpy.matrix=None,
tol: float=0.00001) -> numpy.matrix:
"""
creates a matrix who's respective rows sum to the row_totals and who's respective columns sum to the column_totals
within the tolerance.
:param row_totals:
:param column_totals:
:param maximum_iterations: big number to stop this running forever
:param tol: how close is close enough?
:return:
"""
if not a:
a = numpy.matrix(numpy.ones(shape=(row_totals.shape[0], column_totals.shape[1])))
e = numpy.matrix(numpy.ones(shape=row_totals.shape))
for _ in range(maximum_iterations):
r_hat = row_scaling(a, row_totals, e)
a = matrix_multiply(r_hat, a)
s_hat = column_scaling(a, column_totals, e)
a = matrix_multiply(a, s_hat)
if is_matrix_close_to_i(r_hat, tol) and is_matrix_close_to_i(s_hat, tol):
return a
return a
def run_model(consumption: numpy.matrix, production: numpy.matrix,
emissions: numpy.matrix, final_demand: numpy.matrix=None) -> numpy.matrix:
"""
:param consumption: U
:param production: V
:param emissions: e
:return:
"""
i = numpy.identity(len(consumption))
q = get_row_sum(production.source_data.elements)
g = get_col_sum(production.source_data.elements)
if final_demand is None:
y = get_row_sum(consumption.source_data.elements)
else:
y = final_demand
b = matrix_divide(consumption.source_data.elements, diagonal(g))
d = matrix_divide(production.source_data.elements, diagonal(q))
a = matrix_multiply(b, d)
money_part = matrix_multiply(inv(i - a), y)
return element_wise_divide(emissions.source_data.elements, money_part)
def row_scaling(a: numpy.matrix,
row_totals: numpy.matrix,
e: numpy.matrix) -> numpy.matrix:
r_hat = matrix_divide(diagonal(row_totals),
diagonal(matrix_multiply(a, e)))
return r_hat
def column_scaling(a: numpy.matrix,
column_totals: numpy.matrix,
e: numpy.matrix) -> numpy.matrix:
s_hat = matrix_divide(diagonal(column_totals),
diagonal(matrix_multiply(e.T, a)))
return s_hat
def test_multi_dimension_float(self):
a = numpy.matrix([[3.0], [4.0]])
b = numpy.matrix([1.0, 2.0])
d = matrix_multiply(a, b)
numpy.testing.assert_allclose(d, [[3.0, 6.0], [4.0, 8.0]])
def test_multi_dimension(self):
a = numpy.matrix([[3], [4]])
b = numpy.matrix([1, 2])
d = matrix_multiply(a, b)
numpy.testing.assert_allclose(d, [[3, 6], [4, 8]])
def test_simple_float(self):
a = numpy.matrix([1.0, 2.0])
b = numpy.matrix([[3.0], [4.0]])
d = matrix_multiply(a, b)
numpy.testing.assert_allclose(d, [[11.0]])
def test_simple_case(self):
a = numpy.matrix([1, 2])
b = numpy.matrix([[3], [4]])
d = matrix_multiply(a, b)
numpy.testing.assert_allclose(d, [[11]])
def test_type(self):
a = numpy.matrix([[3], [4]])
b = numpy.matrix([1, 2])
d = matrix_multiply(a, b)
self.assertIs(type(d), numpy.matrix)
