class TestArray(MLTestCase):
def setUp(self):
self.x = Array([1, 2, 3])
def test__check(self):
with self.assertRaises(AssertionError):
self.x + Array([1, 2])
def test_addition(self):
self.assertEqual(self.x + Array([1, 1, 1]), Array([2, 3, 4]))
def test_subtraction(self):
self.assertEqual(self.x - self.x, Array([0, 0, 0]))
def test_multiplication(self):
self.assertEqual(2 * self.x, Array([2, 4, 6]))
self.assertEqual(self.x * -1, Array([-1, -2, -3]))
def test_neg(self):
self.assertEqual(-self.x + self.x, Array([0, 0, 0]))
def test_dot(self):
self.assertEqual(self.x.dot(Array([2, 4, 6])), 1 * 2 + 2 * 4 + 3 * 6)
def test_norm(self):
self.assertEqual(self.x.norm(), (1**2 + 2**2 + 3**2)**0.5)
def test_get_set_item(self):
x = Array([1, 1, 1])
self.assertEqual(x[2], 1)
x[2] = 8
self.assertEqual(x[2], 8)
def test_no_noise(self):
test_X = [Array([1, j, j**2]) for j in range(10)]
weights = Array([1, 0.2, 1])
test_y = Array([weights.dot(x) for x in test_X])
reg = LinearRegression()
reg.fit(test_X, test_y)
self.assertArrayEqual(reg.coef_, weights, 2)
class TestFunction(MLTestCase):
def setUp(self):
self.x = Array([1, 2, 3])
self.f = Function(lambda x: x.norm())
def test_call(self):
self.assertEqual(self.f(self.x), self.x.norm())
def test_mul(self):
g = 3 * self.f
h = self.f * -1
self.assertEqual(g(self.x), 3 * self.x.norm())
self.assertEqual(h(self.x), -self.x.norm())
def test_add(self):
double_norm = self.f + self.f
self.assertEqual(double_norm(self.x), 2 * self.x.norm())
def test_sub(self):
self.assertEqual((self.f - self.f)(self.x), 0)
def test_grad(self):
# D(||x||) = x / ||x||
df = Function(lambda x: x.norm()).grad(self.x)
expected = self.x / self.x.norm()
self.assertArrayEqual(df, expected)
# D(||x||^2) = 2 * x
df = Function(lambda x: x.norm()**2).grad(self.x)
expected = 2 * self.x
self.assertArrayEqual(df, expected)
# D(x[0]^2 - x[1] * x[2]) = (2 * x[0], -x[2], -x[1])
df = Function(lambda x: x[0]**2 - x[1] * x[2]).grad(self.x)
expected = Array([2 * self.x[0], -self.x[2], -self.x[1]])
self.assertArrayEqual(df, expected)
def test_minimize(self):
# ||x - v||^2 is minimized when x = v
f = Function(lambda x: (x - Array(range(len(x)))).norm_squared())
for dims in range(1, 3):
self.assertArrayEqual(f.minimize(dims=dims), Array(range(dims)), 3)
def test_grad(self):
# D(||x||) = x / ||x||
df = Function(lambda x: x.norm()).grad(self.x)
expected = self.x / self.x.norm()
self.assertArrayEqual(df, expected)
# D(||x||^2) = 2 * x
df = Function(lambda x: x.norm()**2).grad(self.x)
expected = 2 * self.x
self.assertArrayEqual(df, expected)
# D(x[0]^2 - x[1] * x[2]) = (2 * x[0], -x[2], -x[1])
df = Function(lambda x: x[0]**2 - x[1] * x[2]).grad(self.x)
expected = Array([2 * self.x[0], -self.x[2], -self.x[1]])
self.assertArrayEqual(df, expected)
def setUp(self):
self.x = Array([1, 2, 3])
def test_get_set_item(self):
x = Array([1, 1, 1])
self.assertEqual(x[2], 1)
x[2] = 8
self.assertEqual(x[2], 8)
def test_dot(self):
self.assertEqual(self.x.dot(Array([2, 4, 6])), 1 * 2 + 2 * 4 + 3 * 6)
def test_neg(self):
self.assertEqual(-self.x + self.x, Array([0, 0, 0]))
def test_multiplication(self):
self.assertEqual(2 * self.x, Array([2, 4, 6]))
self.assertEqual(self.x * -1, Array([-1, -2, -3]))
def test_addition(self):
self.assertEqual(self.x + Array([1, 1, 1]), Array([2, 3, 4]))
def squared_error(w):
Xw = Array([x.dot(w) for x in X])
return (Xw - y).norm()
def transform(self, X):
if self.coef_ is None:
raise NotImplementedError("Must fit the model first")
return Array([1 / (1 + exp(-self.coef_.dot(x))) for x in X])
def setUp(self):
self.x = Array([1, 2, 3])
self.f = Function(lambda x: x.norm())
def test_minimize(self):
# ||x - v||^2 is minimized when x = v
f = Function(lambda x: (x - Array(range(len(x)))).norm_squared())
for dims in range(1, 3):
self.assertArrayEqual(f.minimize(dims=dims), Array(range(dims)), 3)
def test__check(self):
with self.assertRaises(AssertionError):
self.x + Array([1, 2])
def grad(self, x):
return Array([derivative(self, x, eps(x, j)) for j in range(len(x))])
def test_subtraction(self):
self.assertEqual(self.x - self.x, Array([0, 0, 0]))