def test_expectedTable(self):
# CrossTable:
# [2, 0, 3],
# [2, 1, 0],
table = CrossTable(self.X, self.Y)
tab = table.getExpectedTable()
answer = [[20 / 8.0, 5 / 8.0, 15 / 8.0], [12 / 8.0, 3 / 8.0, 9 / 8.0]]
np.testing.assert_array_equal(answer, tab)
tab = table.getExpectedProbtable()
answer = [[20 / 64.0, 5 / 64.0, 15 / 64.0],
[12 / 64.0, 3 / 64.0, 9 / 64.0]]
np.testing.assert_array_equal(answer, tab)
np.testing.assert_array_equal(table.getProbCols(),
[4.0 / 8, 1.0 / 8, 3.0 / 8])
np.testing.assert_array_equal(table.getProbRows(), [5.0 / 8, 3.0 / 8])
def test_expectedTable(self):
# CrossTable:
# [2, 0, 3],
# [2, 1, 0],
table = CrossTable(self.X, self.Y)
tab = table.getExpectedTable()
answer = [
[20/8.0, 5/8.0, 15/8.0],
[12/8.0, 3/8.0, 9/8.0]
]
np.testing.assert_array_equal(answer, tab)
tab = table.getExpectedProbtable()
answer = [
[20/64.0, 5/64.0, 15/64.0],
[12/64.0, 3/64.0, 9/64.0]
]
np.testing.assert_array_equal(answer, tab)
np.testing.assert_array_equal(table.getProbCols(), [4.0/8, 1.0/8, 3.0/8])
np.testing.assert_array_equal(table.getProbRows(), [5.0/8, 3.0/8])
def cramer(X, Y):
'''
Define Cramer's relationship coefficient of the rasters for discrete values
Coefficient change between [0, 1]
0 - no dependence
1 - full connection
@param X First raster's array
@param Y Second raster's array
'''
table = CrossTable(X, Y)
rows, cols = table.shape
t_expect = table.getExpectedTable()
# Mask T* to prevent division by zero
t_expect = np.ma.array(t_expect, mask=(t_expect == 0))
# chi-square coeff = sum((T-T*)^2/T*)
x2 = np.sum(np.square(table.T - t_expect)/t_expect)
# CRAMER CONTINGENCY COEF. = sqrt(chi-square / (total * min(s-1,r-1)))
# s, r - raster grauations
Cramer = math.sqrt(x2/(table.n*min(cols-1, rows-1)))
return Cramer
