def test_trig1():
Graph = ad.ComputationalGraph()
x = ad.Node(value=math.pi, Graph=Graph)
func = ad.cosr(x) - ad.sinr(2*x) + ad.tanr(x)
Graph.ComputeValue()
Graph.ComputeGradient()
assert abs(func.value - (-1)) < .000001 and abs(x.deri - (-1)) < .000001
def test_exp():
Graph = ad.ComputationalGraph()
x = ad.Node(value=3, Graph=Graph)
func = ad.expr(x)
Graph.ComputeValue()
Graph.ComputeGradient()
assert (func.value, x.deri) == (np.exp(3), np.exp(3))
def test_logsqrt():
Graph = ad.ComputationalGraph()
x = ad.Node(value=2, Graph=Graph)
func = ad.sqrtr(ad.logr(x))
Graph.ComputeValue()
Graph.ComputeGradient()
assert (func.value, x.deri) == (np.sqrt(np.log(2)), .25 * (1 / (np.sqrt(np.log(2)))))
def test_revpower():
Graph = ad.ComputationalGraph()
x = ad.Node(value=2, Graph=Graph)
func = 2 - 2**x
Graph.ComputeValue()
Graph.ComputeGradient()
assert(func.value, x.deri) == (-2, -4*np.log(2))
def test_div():
Graph = ad.ComputationalGraph()
x = ad.Node(value=5, Graph=Graph)
func = 1 / x + x/2
Graph.ComputeValue()
Graph.ComputeGradient()
assert (func.value, x.deri) == (2.7, .46)
def test_addnegexp():
Graph = ad.ComputationalGraph()
x = ad.Node(value=5, Graph=Graph)
func = -x + 2*x**2 + 2
Graph.ComputeValue()
Graph.ComputeGradient()
assert(func.value, x.deri) == (47, 19)
def test_trig3():
Graph = ad.ComputationalGraph()
x = ad.Node(value=0, Graph=Graph)
func = ad.sinhr(x) + ad.coshr(x) + ad.tanhr(x)
Graph.ComputeValue()
Graph.ComputeGradient()
assert (func.value, x.deri) == (1, 2)
def test_trig2():
Graph = ad.ComputationalGraph()
x = ad.Node(value=0, Graph=Graph)
func = ad.asinr(x) * ad.acosr(x**2) + 2*ad.atanr(x)
Graph.ComputeValue()
Graph.ComputeGradient()
assert (func.value, x.deri) == (0, 2+np.arccos(0))
def test_submulti():
Graph = ad.ComputationalGraph()
x = ad.Node(value=3, Graph=Graph)
y = ad.Node(value=2, Graph=Graph)
func = 1 + x*5 - 2*y
Graph.ComputeValue()
Graph.ComputeGradient()
assert(func.value, x.deri, y.deri) == (12, 5, -2)
def test_series():
Graph = ad.ComputationalGraph()
x = ad.Node(value=5, Graph=Graph)
f = x ** 2
C = np.array([[4, 2, 3]])
D = 1
Vals, Ders = Graph.SeriesValues(C, D, Graph)
assert np.array_equal(Vals, [16, 4, 9]) and np.array_equal(Ders, [[8], [4], [6]])
def test_jacobian():
Graph = ad.ComputationalGraph()
x = ad.Node(value=3, Graph=Graph)
y = ad.Node(value=5, Graph=Graph)
z = ad.Node(value=1, Graph=Graph)
f = np.array([-2*x, 2*y+z, 3*x+3*y+2*z])
func = ad.ReverseVecFunc(f, x=x, y=y, z=z)
val, jacobian = func.value(Graph)
assert np.array_equal(val, [-6, 11, 26]) and np.array_equal(jacobian, [[-2, 0, 0], [0, 2, 1], [3, 3, 2]])
def test_jacobian_series():
Graph = ad.ComputationalGraph()
x = ad.Node(value=3, Graph=Graph)
y = ad.Node(value=4, Graph=Graph)
X = [1, 3]
Y = [2, 4]
C = np.array([X, Y])
D = 2 # Dimension
f = np.array([x * 3, 5 * y ** 2])
func = ad.ReverseVecFunc(f, x=x, y=y)
vals, derivs = func.Seriesvalue(C, D, Graph)
assert np.array_equal(vals, [[3, 20], [9, 80]]) and np.array_equal(derivs, [[[3, 0], [0,20]], [[3, 0], [0, 40]]])