def testlogfunction():
# Test if log approximation works as expected
Ptensor = PolynomialTensor()
x = torch.tensor(np.linspace(1, 10, 10), dtype=torch.double)
expected = torch.tensor(
[
3.7160e-04,
6.9319e-01,
1.0986e00,
1.3863e00,
1.6096e00,
1.7932e00,
1.9526e00,
2.1056e00,
2.2835e00,
2.5537e00,
],
dtype=torch.double,
)
result = Ptensor.get_val("log", x.clone())
# allclose function to compare the expected values and approximations with fixed precision
assert torch.allclose(expected,
torch.tensor(result, dtype=torch.double),
atol=1e-03)
def testcustomfunction():
P = PolynomialTensor()
P.addfunction("Custom", 10, [[0, 10, 100, 10, P.fit_function]],
lambda x: x + 2)
assert round(P.get_val("Custom", 3)) == 5
assert round(P.get_val("Custom", 6)) == 8
def testinterpolate():
# Test if interpolation function works as expected by verifying an approximation of exponential function
expected = torch.tensor([1.2220, 2.9582, 7.1763, 20.3064, 54.4606],
dtype=torch.double)
Ptensor = PolynomialTensor()
f1 = Ptensor.interpolate((lambda x: np.exp(x)), np.linspace(0, 10, 50))
assert torch.allclose(expected,
torch.tensor(f1(torch.tensor([0, 1, 2, 3, 4]))),
1e-04)
def testSigmoid():
Ptensor = PolynomialTensor()
x = torch.tensor(np.linspace(-3, 3, 10), dtype=torch.double)
expected = torch.tensor(
[
0.0337, 0.0886, 0.1759, 0.2921, 0.4283, 0.5717, 0.7079, 0.8241,
0.9114, 0.9663
],
dtype=torch.double,
)
# allclose function to compare the expected values and approximations with fixed precision
result = Ptensor.get_val("sigmoid", x)
assert torch.allclose(expected, result, atol=1e-03)
def testExp():
Ptensor = PolynomialTensor()
x = torch.tensor(np.linspace(-3, 3, 10), dtype=torch.double)
expected = torch.tensor(
[
0.0498, 0.0970, 0.1889, 0.3679, 0.7165, 1.1176, 2.9503, 5.1088,
10.2501, 20.2955
],
dtype=torch.double,
)
# allclose function to compare the expected values and approximations with fixed precision
result = Ptensor.get_val("exp", x)
assert torch.allclose(expected, result, atol=1e-03)
def testsigmoidtaylor():
expected = torch.tensor(
[
0.1000, 0.1706, 0.2473, 0.3392, 0.4447, 0.5553, 0.6608, 0.7527,
0.8294, 0.9000
],
dtype=torch.double,
)
P = PolynomialTensor()
x = torch.tensor(np.linspace(-2, 2, 10), dtype=torch.double)
result = P.sigmoid(x)
# allclose function to compare the expected values and approximations with fixed precision
assert torch.allclose(expected,
torch.tensor(result, dtype=torch.double),
atol=1e-03)
def testexptaylor():
expected = torch.tensor(
[
-0.1076, 0.0664, 0.1852, 0.3677, 0.7165, 1.3956, 2.7180, 5.2867,
10.2325, 19.5933
],
dtype=torch.double,
)
P = PolynomialTensor()
x = torch.tensor(np.linspace(-3, 3, 10), dtype=torch.double)
result = P.exp(x)
# allclose function to compare the expected values and approximations with fixed precision
assert torch.allclose(expected,
torch.tensor(result, dtype=torch.double),
atol=1e-03)
def testtanh():
# Test if tanh approximation works as expected
Ptensor = PolynomialTensor()
x = torch.tensor(np.linspace(-3, 3, 10), dtype=torch.double)
expected = torch.tensor(
[
-0.9937, -0.9811, -0.9329, -0.7596, -0.3239, 0.3239, 0.7596,
0.9329, 0.9811, 0.9937
],
dtype=torch.double,
)
result = Ptensor.get_val("tanh", x)
# allclose function to compare the expected values and approximations with fixed precision
assert torch.allclose(expected,
torch.tensor(result, dtype=torch.double),
atol=1e-03)
def testrandomfunction():
x = torch.tensor(np.linspace(-3, 3, 10), dtype=torch.double)
P = PolynomialTensor(function=lambda x: x * 2)
scaled_result = P.get_val("exp", x.clone())
Ptensor = PolynomialTensor()
original_result = Ptensor.get_val("exp", x)
assert torch.all(torch.eq(scaled_result, torch.mul(original_result, 2)))