def forward_propagation(self, turn):
"""
前馈计算
:param turn: 计算第turn个样本数据
:return:
"""
self.i = self.x[turn]
self.h = sigmoid(np.dot(self.W1, self.i))
self.o = sigmoid(np.dot(self.W2, self.h))
def test(self, file_name):
"""
测试
:return:
"""
self.read_data(file_name)
count = 0
for i in range(self.n):
h = sigmoid(np.dot(self.W1, self.x[i]))
o = sigmoid(np.dot(self.W2, h))
predict = list(o).index(max(list(o)))
if predict == i:
count += 1
# print count * 1.0 / 10, '测试集的损失函数是：',
# print self.loss_function()
return count * 1.0 / 10, self.loss_function()
def loss_function(self):
"""
损失函数
:return:
"""
self.loss = 0
for i in range(self.n):
h = sigmoid(np.dot(self.W1, self.x[i]))
o = sigmoid(np.dot(self.W2, h))
for j in range(self.output_layer_size):
self.loss += (o[j] - self.y[i][j]) ** 2
self.loss *= 0.5
# 先计算W1的权值
for i in range(self.hidden_layer_size):
for j in range(self.input_layer_size):
self.loss += self.lambda_ * self.W1[i][j] * self.W1[i][j] / self.n * 0.5
# 再计算W2的权值
for i in range(self.output_layer_size):
for j in range(self.hidden_layer_size):
self.loss += self.lambda_ * self.W2[i][j] * self.W2[i][j] / self.n * 0.5
return self.loss
def feedforward(self):
# feedforward(forwardpropagation) we will predict outputs
self.layer1 = sigmoid(np.dot(self.input, self.weights1))
self.layer2 = sigmoid(np.dot(self.layer1, self.weights2))
self.output = sigmoid(np.dot(self.layer2, self.weights3))
