def test_backprop_1500_iteration(self):
X = np.array([[0.05, 0.10]])
y = np.array([[0.01, 0.99]])
nn = MyNeuralNet()
nn.input = np.array(list(zip(X, y)))
self.setup_test_conf(nn)
nn.learn(1500, 0.5, 1)
self.assertEqual(nn.weights[0][0][0], 0.1747576240731027)
self.assertEqual(nn.weights[0][0][1], 0.24951524814620638)
self.assertEqual(nn.weights[0][1][0], 0.2740253089080296)
self.assertEqual(nn.weights[0][1][1], 0.3480506178160598)
self.assertEqual(nn.output.shape, (1, 2))
def test_backprop_1_iteration(self):
X = np.array([[0.05, 0.10]])
y = np.array([[0.01, 0.99]])
nn = MyNeuralNet()
nn.input = np.array(list(zip(X, y)))
self.setup_test_conf(nn)
nn.learn(1, 0.5, 1)
self.assertEqual(nn.weights[0][0][0], 0.14978071613276281)
self.assertEqual(nn.weights[0][0][1], 0.19956143226552567)
self.assertEqual(nn.weights[0][1][0], 0.24975114363236958)
self.assertEqual(nn.weights[0][1][1], 0.29950228726473915)
self.assertEqual(nn.weights[1][0][0], 0.35891647971788465)
self.assertEqual(nn.weights[1][0][1], 0.4086661860762334)
self.assertEqual(nn.weights[1][1][0], 0.5113012702387375)
self.assertEqual(nn.weights[1][1][1], 0.5613701211079891)
def test_relu(self):
# XOR
X = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
y = np.array([[0, 1], [1, 0], [1, 0], [0, 1]])
nn = MyNeuralNet()
nn.input = np.array(list(zip(X, y)))
nn.add_layer(100)
nn.add_layer(100)
nn.add_layer(2)
nn.learn(2000, 0.01, 1, activation='relu')
res = nn.classify(X)
print(res)
self.assertEqual(int(round(res[0][0])), 0)
self.assertEqual(int(round(res[0][1])), 1)
self.assertEqual(int(round(res[1][0])), 1)
self.assertEqual(int(round(res[1][1])), 0)
def test_classify(self):
X = np.array([[0, 0], [0, 1], [1, 0], [1, 1], [0, 0]])
y = np.array([[0], [1], [1], [0], [0]])
nn = MyNeuralNet()
nn.input = np.array(list(zip(X, y)))
nn.add_layer(4, 0)
nn.add_layer(1, 0)
nn.learn(1500, 10, 1)
res = nn.classify(np.array([[0, 0], [1, 0]]))
self.assertEqual(int(round(res[0][0])), 0)
self.assertEqual(int(round(res[1][0])), 1)
def test_multiple_input_50_neuron_2_layer(self):
X = np.array([[0, 0], [0, 1], [1, 0], [1, 1], [0, 0]])
y = np.array([[0], [1], [1], [0], [0]])
nn = MyNeuralNet()
nn.input = np.array(list(zip(X, y)))
nn.add_layer(4, 0)
nn.add_layer(1, 0)
nn.learn(1500, 1, 5)
res = nn.classify([[0, 0], [0, 1], [1, 0], [1, 1], [0, 0]])
self.assertEqual(list(map(lambda x: int(round(x[0])), res)),
[0, 1, 1, 0, 0])
def test_raise_illegal_output(self):
X = np.array([[0, 0, 1], [0, 1, 1], [1, 0, 1], [1, 1, 1], [0, 0, 0]])
y = np.array([[0, 0], [1, 1], [1, 1], [0, 0], [0, 0]])
nn = MyNeuralNet()
nn.input = np.array(list(zip(X, y)))
nn.add_layer(1, 0)
with self.assertRaises(ValueError):
nn.learn(1, 1, 1)
def test_simple_net(self):
X = np.array([[0.05]])
y = np.array([[0.01]])
nn = MyNeuralNet()
nn.input = np.array(list(zip(X, y)))
nn.add_layer(1, 0)
nn.learn(1500, 0.3, 1)
self.assertEqual(len(nn.weights), 1)
self.assertEqual(nn.weights[0].shape, (1, 2))
self.assertEqual(nn.weights[0][0][0], 1.60050105653771)
def test_gradient_checking(self):
X = np.array([[0.05, 0.10]])
y = np.array([[0.01, 0.99]])
nn = MyNeuralNet()
nn.activation = 'relu'
nn.input = np.array(list(zip(X, y)))
self.setup_test_conf(nn)
grad_approx = np.array(self.calc_gradients(nn))
nn.feed_forward(nn.input[:, 0])
grad = np.array(nn.backward(nn.input[:, 1]))
self.assertTrue(np.allclose(grad, grad_approx))
def test_raise_no_layers(self):
nn = MyNeuralNet()
with self.assertRaises(ValueError):
nn.learn(1, 0.3, 1)
nn.last_layer_neuron_count