def get_selection_S(F):
"""
The F and S notation is from Yang and Nielsen 2008.
@param F: a selection value for each codon, up to an additive constant
@return: selection differences F_j - F_i, also known as S_ij
"""
e = algopy.ones_like(F)
return algopy.outer(e, F) - algopy.outer(F, e)
def get_selection_S(F):
"""
The F and S notation is from Yang and Nielsen 2008.
@param F: a selection value for each codon, up to an additive constant
@return: selection differences F_j - F_i, also known as S_ij
"""
e = algopy.ones_like(F)
return algopy.outer(e, F) - algopy.outer(F, e)
def get_selection_S(F):
"""
The F and S notation is from Yang and Nielsen 2008.
Speed matters.
@param F: a selection value for each codon, up to an additive constant
@return: selection differences F_j - F_i, also known as S_ij
"""
# FIXME: use algopy.ones_like when it becomes available
e = algopy.ones_like(F)
# # FIXME: instead of the following block
# e = algopy.zeros_like(F)
# for i in range(F.shape[0]):
# e[i] = 1.
return algopy.outer(e, F) - algopy.outer(F, e)
def test_outer(self):
x = numpy.arange(4)
cg = algopy.CGraph()
x = algopy.Function(x)
x1 = x[:x.size//2]
x2 = x[x.size//2:]
y = algopy.trace(algopy.outer(x1,x2))
cg.trace_off()
cg.independentFunctionList = [x]
cg.dependentFunctionList = [y]
cg2 = algopy.CGraph()
x = algopy.Function(x)
x1 = x[:x.size//2]
x2 = x[x.size//2:]
z = algopy.dot(x1,x2)
cg2.trace_off()
cg2.independentFunctionList = [x]
cg2.dependentFunctionList = [z]
assert_array_almost_equal(cg.jacobian(numpy.arange(4)), cg2.jacobian(numpy.arange(4)))
