def test_reg_grad_data2_2(self):
y = self.data2[:, -1:]
X = self.data2[:, :-1]
m, n = X.shape
intercept = ones((m, 1), dtype=float64)
X = append(intercept, X, axis=1)
theta = ones((n + 1, 1), dtype=float64)
_lambda = 1000000
assert_allclose([[-338407.808], [-759558338.468], [-1092403.298]],
reg_grad(X, y, theta, _lambda),
rtol=0, atol=0.001)
def test_reg_grad_data2_1(self):
y = self.data2[:, -1:]
X = self.data2[:, :-1]
m, n = X.shape
intercept = ones((m, 1), dtype=float64)
X = append(intercept, X, axis=1)
theta = ones((n + 1, 1), dtype=float64)
_lambda = 0
assert_allclose([[-338407.808], [-759579615.064], [-1113679.894]],
reg_grad(X, y, theta, _lambda),
rtol=0, atol=0.001)
def test_reg_grad_data1_3(self):
y = self.data1[:, -1:]
X = self.data1[:, :-1]
m, n = X.shape
intercept = ones((m, 1), dtype=float64)
X = append(intercept, X, axis=1)
theta = array([[-1], [2]])
_lambda = 750
assert_allclose([[9.480465], [104.783153]],
reg_grad(X, y, theta, _lambda),
rtol=0, atol=0.001)
def test_reg_grad_data1_2(self):
y = self.data1[:, -1:]
X = self.data1[:, :-1]
m, n = X.shape
intercept = ones((m, 1), dtype=float64)
X = append(intercept, X, axis=1)
theta = ones((n + 1, 1), dtype=float64)
_lambda = 100
assert_allclose([[3.320665], [25.265821]],
reg_grad(X, y, theta, _lambda),
rtol=0, atol=0.001)
def test_reg_grad_data2_3(self):
y = self.data2[:, -1:]
X = self.data2[:, :-1]
m, n = X.shape
intercept = ones((m, 1), dtype=float64)
X = append(intercept, X, axis=1)
theta = array([[-25.3], [32], [7.8]])
_lambda = 1000000
assert_allclose([[-276391.444681],
[-615660007.370213],
[-740838.968085]],
reg_grad(X, y, theta, _lambda),
rtol=0, atol=0.001)
def test_reg_grad_data1_6(self):
y = self.data1[:, -1:]
X = self.data1[:, :-1]
m, n = X.shape
intercept = ones((m, 1), dtype=float64)
X = append(intercept, X, axis=1)
theta = array([[-12.4], [23.56]])
_lambda = 943
def J(theta):
return reg_cost_func(X, y, theta, _lambda)
assert_allclose(reg_grad(X, y, theta, _lambda),
numerical_grad(J, theta, self.err),
rtol=0, atol=0.001)
def test_reg_grad_data1_4(self):
y = self.data1[:, -1:]
X = self.data1[:, :-1]
m, n = X.shape
intercept = ones((m, 1), dtype=float64)
X = append(intercept, X, axis=1)
theta = -8.4 * ones((n + 1, 1), dtype=float64)
_lambda = 0.762
def J(theta):
return reg_cost_func(X, y, theta, _lambda)
assert_allclose(reg_grad(X, y, theta, _lambda),
numerical_grad(J, theta, self.err),
rtol=0, atol=0.001)