今天,接着跟大家分享一波ReLU函数。这个函数也是激活函数。表达式如下:
y = {
x,x>0
0,x<=0
}
代码实现如下:
def ReLU(inx):
return np.maximum(0,inx)
绘图:
Inx = np.arange(-5,5,0.5)
y = ReLU(Inx)
plt.plot(Inx,y)
plt.ylim(-0.5,5)
plt.show()
可以从图像看出ReLU函数是非线性函数,相比阶跃函数,它的输出结果有0和自身的输入。阶跃函数的输出结果只有0和1。
多维数组运算:
def Matrix_calc():
Matrix1 = np.array([[1,2],[3,4],[5,6]])
dimension = Matrix1.ndim
print('{}的维度:'.format(Matrix1),dimension)
shape = Matrix1.shape
print('{}的形状:'.format(Matrix1),shape)
Matrix2 = np.array([[7,8],[9,0]])
dimension2 = Matrix2.ndim
print('{}的维度:'.format(Matrix2),dimension2)
shape2 = Matrix2.shape
print('{}的形状:'.format(Matrix2),shape2)
Matrix3 = np.dot(Matrix2,Matrix2)#矩阵的点积
print(Matrix3)
print("{}的形状:".format(Matrix3),Matrix3.shape)
dimension3 = Matrix3.ndim
print("{}的维度:".format(Matrix3),dimension3)
运行结果:
构建三层网络:
def network_build(inx,weights,theta):
b = -theta#偏置等于阈值的负值
y = np.dot(inx,weights) + b#输入与权重的点积加上偏置
#等同于 y = inx * weights + b
return y
#第一层网络
inx = np.array([0.5,1])#传入的数据
#print(inx.shape)
weights = np.array([[0.3,0.5,0.7],[0.4,0.6,0.8]])#第一层网络的权重
# print(weights.shape)
# result = np.dot(inx,weights)
# print(result)
theta = np.array([0.1,0.3,0.5])#第一层网络的阈值
# for n in range(5):
# theta = np.array([0.1, 0.3, 0.5])
# weights = np.array([[0.3, 0.5, 0.7], [0.4, 0.6, 0.8]]) # 权重
result = network_build(inx,weights,theta)
output = sigmoid(result)#调用sigmoid函数优化
print(output)#第一层网络的输出结果
print()
print()
#第二层网络
inx2 = output#下层网络的输入数据等于上层网络输出的结果
weights2 = np.array([[0.1,0.3],[0.5,0.7],[0.2,0.4]])#第二层网络的权重
theta2 = np.array([0.1,0.2])#第二层网络的阈值
result2 = network_build(inx2,weights2,theta2)
output2 = sigmoid(result2)
print(output2)
#第三层网络
print()
print()
inx3 = output2
weights3 = np.array([[0.1,0.3,0.5],[0.2,0.4,0.6]])
theta3 = np.array([0.1,0.2,0.3])
y = network_build(inx3,weights3,theta3)
print(y)
output3 = identity_func(y)#调用恒等函数作为激活函数
print(output3)
运行结果:
恒等函数构建:
def identity_func(inx):#恒等函数,输入等于输出,激活函数
return inx
我们结合上面构建三层网络的例子,再做一下调整,也能得到三层网络:
def init_network():#初始化网络
#这个网络中有我们每一层需要的权重以及theta阈值。
network = {}
network['W1'] = np.array([[0.1,0.2,0.3],[0.4,0.5,0.6]])
network['W2'] = np.array([[0.2,0.4],[0.1,0.3],[0.5,0.7]])
network['W3'] = np.array([[0.1,0.3],[0.2,0.4]])
network['theta1'] = np.array([0.1,0.2,0.3])
network['theta2'] = np.array([0.1,0.3])
network['theta3'] = np.array([0.2,0.4])
return network
def forward(network,inx):
b1 = -network['theta1']#偏置1
b2 = -network['theta2']#偏置2
b3 = -network['theta3']#偏置3
result1 = np.dot(inx,network['W1'])+b1
output1 = sigmoid(result1)#调用sigmoid激活函数
print('Sigmoid激活函数输出结果:',output1)
result2 = np.dot(output1,network['W2'])+b2
output2 = ReLU(result2)#调用ReLU激活函数
print('ReLU激活函数输出结果:',output2)
result3 = np.dot(output2, network['W3']) + b3
output3 = identity_func(result3)#调用恒等激活函数
print('identity_func恒等激活函数输出结果:',output3)
运行代码:
network = init_network()
x = [0.1,0.6]#输入的数据
forward(network,x)
运行结果:
最后,感谢大家前来观看鄙人的文章,文中或有诸多不妥之处,还望指出和海涵。



