def test_predict(self):
y_pred = self.clf.predict_proba(self.X)
expected = [Perceptron.sigmoid(0)] * 2
self.assertListEqual(y_pred.tolist(), expected)
y_pred_prime = self.clf.predict_proba(self.X, True)
expected = [Perceptron.sigmoid_prime(0)] * 2
self.assertListEqual(y_pred_prime.tolist(), expected)
def test_sigmoid_prime(self):
subjects = np.array([-2, -1, 0, 1])
expected = np.array([sp(x) for x in subjects])
result = Perceptron.sigmoid_prime(subjects)
self.assertEqual(type(result), np.ndarray)
for r, e in zip(result, expected):
self.assertAlmostEqual(r, e)
def test_sigmoid(self):
subjects = np.array([-2, -1, 0, 1])
expected = np.array([
.1192029220221175559403, .2689414213699951207488, .5,
.7310585786300048792512
])
result = Perceptron.sigmoid(subjects)
self.assertEqual(type(result), np.ndarray)
for r, e in zip(result, expected):
self.assertAlmostEqual(r, e)
def test_grad_w(self):
result_0 = self.clf.grad_w(self.X, self.y)
expected_0 = [
-sum([
-.5 * Perceptron.sigmoid_prime(0) * 1,
.5 * Perceptron.sigmoid_prime(0) * 3
]) / 2, -sum([
-.5 * Perceptron.sigmoid_prime(0) * 2,
.5 * Perceptron.sigmoid_prime(0) * 2
]) / 2, -sum([
-.5 * Perceptron.sigmoid_prime(0) * 3,
.5 * Perceptron.sigmoid_prime(0) * 1
]) / 2
]
self.assertListEqual(result_0.tolist(), expected_0)
def test_grad_b(self):
result_0 = self.clf.grad_b(self.X, self.y)
expected_0 = -sum([
-.5 * Perceptron.sigmoid_prime(0), .5 * Perceptron.sigmoid_prime(0)
]) / 2
self.assertEqual(result_0, expected_0)
def setUp(self):
self.X = np.array([[1., 2., 3.], [3., 2., 1.]])
self.y = np.array([[0, 1]])
self.clf = Perceptron(3)
class TestPerceptron(TestCase):
def setUp(self):
self.X = np.array([[1., 2., 3.], [3., 2., 1.]])
self.y = np.array([[0, 1]])
self.clf = Perceptron(3)
def test_grad_w(self):
result_0 = self.clf.grad_w(self.X, self.y)
expected_0 = [
-sum([
-.5 * Perceptron.sigmoid_prime(0) * 1,
.5 * Perceptron.sigmoid_prime(0) * 3
]) / 2, -sum([
-.5 * Perceptron.sigmoid_prime(0) * 2,
.5 * Perceptron.sigmoid_prime(0) * 2
]) / 2, -sum([
-.5 * Perceptron.sigmoid_prime(0) * 3,
.5 * Perceptron.sigmoid_prime(0) * 1
]) / 2
]
self.assertListEqual(result_0.tolist(), expected_0)
def test_grad_b(self):
result_0 = self.clf.grad_b(self.X, self.y)
expected_0 = -sum([
-.5 * Perceptron.sigmoid_prime(0), .5 * Perceptron.sigmoid_prime(0)
]) / 2
self.assertEqual(result_0, expected_0)
def test_sigmoid(self):
subjects = np.array([-2, -1, 0, 1])
expected = np.array([
.1192029220221175559403, .2689414213699951207488, .5,
.7310585786300048792512
])
result = Perceptron.sigmoid(subjects)
self.assertEqual(type(result), np.ndarray)
for r, e in zip(result, expected):
self.assertAlmostEqual(r, e)
def test_sigmoid_prime(self):
subjects = np.array([-2, -1, 0, 1])
expected = np.array([sp(x) for x in subjects])
result = Perceptron.sigmoid_prime(subjects)
self.assertEqual(type(result), np.ndarray)
for r, e in zip(result, expected):
self.assertAlmostEqual(r, e)
def test_predict(self):
y_pred = self.clf.predict_proba(self.X)
expected = [Perceptron.sigmoid(0)] * 2
self.assertListEqual(y_pred.tolist(), expected)
y_pred_prime = self.clf.predict_proba(self.X, True)
expected = [Perceptron.sigmoid_prime(0)] * 2
self.assertListEqual(y_pred_prime.tolist(), expected)
def test_cost(self):
cost = self.clf.cost(self.X, self.y)
predicted = self.clf.predict_proba(self.X)
expected = sum([(0. - a)**2 for a in predicted]) / 4
self.assertEqual(cost, expected)