【theano】05-classification分类例子

1.导入包

2.

#正确率方法定义
def compute_accuracy(y_target, y_predict):    
    correct_prediction = np.equal(y_predict, y_target)    
    accuracy = np.sum(correct_prediction)/len(correct_prediction)
    return accuracy

3.数据

#可看成rng是np.random的简写
rng = np.random


N = 400 # training sample size数据的example
feats = 784 # number of input variables数据点

# generate a dataset: D = (input_values, target_class)    #输入数据、类别
D = (rng.randn(N, feats), rng.randint(size=N, low=0, high=2))    #生成虚拟数据 类别：0,1

4.

# Declare Theano symbolic variables

x = T.dmatrix("x")
y = T.dvector("y")

5.权重偏置

# initialize the weights and biases

W = theano.shared(rng.randn(feats), name="w")    #Weights
b = theano.shared(0.1, name="b")    #biases

6.loss

#分类层
p_1 = T.nnet.sigmoid(T.dot(x, W) + b) # Logistic Probability that target = 1 (activation function)
prediction = p_1 > 0.5 # The prediction thresholded    >0.5被认为是1

#计算loss
xent = -y * T.log(p_1) - (1-y) * T.log(1-p_1) # Cross-entropy loss function
# or
# xent = T.nnet.binary_crossentropy(p_1, y) # this is provided by theano

cost = xent.mean() + 0.01 * (W ** 2).sum()    # The cost to minimize (l2 regularization)
#L1loss：0.01 * (W ** 2).sum()为了避免overfitting(过拟合)

#求梯度
gW, gb = T.grad(cost, [W, b]) # Compute the gradient of the cost

7.compile

learning_rate = 0.1    #学习率
train = theano.function(
    inputs=[x, y],
    outputs=[prediction, xent.mean()],    #xent.mean()是平均损失
    updates=((W, W - learning_rate * gW), (b, b - learning_rate * gb)))    #更新权重偏置

predict = theano.function(inputs=[x], outputs=prediction)    #预测

8.training

for i in range(500):
    pred, err = train(D[0], D[1])    #D[0]数据 D[1]类别 train返回的一个是predict一个是loss
    if i % 50 == 0:
        print('cost:', err)
        print("accuracy:", compute_accuracy(D[1], predict(D[0])))    #计算准确率

print("target values for D:")
print(D[1])
print("prediction on D:")
print(predict(D[0]))