class MLP:
def __init__(self, input_size, hidden_size_1, hidden_size_2, output_size):
self.input_layer = InputLayer(input_size, hidden_size_1, ReLU)
self.hidden_layer = HiddenLayer(hidden_size_1, hidden_size_2)
self.output_layer = SoftmaxOutputLayer(hidden_size_2, output_size)
def predict(self, x):
x = self.input_layer.forward(x)
x = self.hidden_layer.forward(x)
prob = self.output_layer.predict(x)
pred = np.argmax(prob, axis=-1)
return pred
def loss(self, x, y):
x = self.input_layer.forward(x)
x = self.hidden_layer.forward(x)
loss = self.output_layer.forward(x, y)
return loss
def gradient(self):
d_prev = 1
d_prev = self.output_layer.backward(d_prev=d_prev)
d_prev = self.hidden_layer.backward(d_prev)
self.input_layer.backward(d_prev)
def update(self, learning_rate):
self.input_layer.W -= self.input_layer.dW * learning_rate
self.input_layer.b -= self.input_layer.db * learning_rate
self.hidden_layer.W -= self.hidden_layer.dW * learning_rate
self.hidden_layer.b -= self.hidden_layer.db * learning_rate
self.output_layer.W -= self.output_layer.dW * learning_rate
self.output_layer.b -= self.output_layer.db * learning_rate
def __init__(self, input_size, hidden_size_1, hidden_size_2, output_size):
self.input_layer = InputLayer(input_size, hidden_size_1, ReLU)
self.hidden_layer = HiddenLayer(hidden_size_1, hidden_size_2)
self.output_layer = SoftmaxOutputLayer(hidden_size_2, output_size)
[[0.03511903 0.25949646 0.70538451]
[0.00108758 0.00270191 0.99621051]]
Forward:
4.631404434078592
Backward:
[[ 0.25645564 0.50146461 1.92169975]
[ 2.7215362 -0.314622 10.96489546]]
dW:
[[-5.43791927e-04 4.98649045e-01 -4.98105254e-01]
[ 5.43791927e-04 -4.98649045e-01 4.98105254e-01]
[ 2.71895963e-04 -2.49324523e-01 2.49052627e-01]]
db:
[-0.48189669 -0.36890082 0.85079751]
"""
softmax_output_layer = SoftmaxOutputLayer(3, 3)
softmax_output_layer.W = np.array([[1.12, -2.34, 3.12], [-1, 0.5, -0.13],
[1.1, -11, 11]])
softmax_output_layer.b = np.array([-1, 1, 2])
temp_y = np.array([[0, 1, 0], [1, 0, 0], [0, 0, 1]])
temp_y_hat = np.array([[0.5, 0.0, 0.1], [0.3, 0.2, 2.1], [3.5, 1.0, 1.1]])
print('Cross-entropy loss: \n',
softmax_output_layer.ce_loss(temp_y_hat, temp_y))
temp8 = np.array([[0, 0, 0], [-1, 1, 0.5]])
print('Predict: \n', softmax_output_layer.predict(temp8))
temp_y = np.array([[1, 0, 0], [0, 1, 0]])
print('Forward: \n', softmax_output_layer.forward(temp8, temp_y))
print('Backward: \n', softmax_output_layer.backward(d_prev=1))