《神经网络与深度学习：案例与实践》动手练习1.3

飞桨AI Studio星河社区-人工智能学习与实训社区

动手练习1.3

执行上述算子的反向过程，并验证梯度是否正确。

import math
 
 
class Op(object):
    def __init__(self):
        pass
 
    def __call__(self, inputs):
        return self.forward(inputs)
 
    # 前向函数
    # 输入：张量inputs
    # 输出：张量outputs
    def forward(self, inputs):
        # return outputs
        raise NotImplementedError
 
    # 反向函数
    # 输入：最终输出对outputs的梯度outputs_grads
    # 输出：最终输出对inputs的梯度inputs_grads
    def backward(self, outputs_grads):
        # return inputs_grads
        raise NotImplementedError
 
 
class add(Op):
    def __init__(self):
        super(add, self).__init__()
 
    def __call__(self, x, y):
        return self.forward(x, y)
 
    def forward(self, x, y):
        self.x = x
        self.y = y
        outputs = x + y
        return outputs
 
    def backward(self, grads):
        grads_x = grads * 1
        grads_y = grads * 1
        return grads_x, grads_y
 
 
class multiply(Op):
    def __init__(self):
        super(multiply, self).__init__()
 
    def __call__(self, x, y):
        return self.forward(x, y)
 
    def forward(self, x, y):
        self.x = x
        self.y = y
        outputs = x * y
        return outputs
 
    def backward(self, grads):
        grads_x = grads * self.y
        grads_y = grads * self.x
        return grads_x, grads_y
 
 
class exponential(Op):
    def __init__(self):
        super(exponential, self).__init__()
 
    def forward(self, x):
        self.x = x
        outputs = math.exp(x)
        return outputs
 
    def backward(self, grads):
        grads = grads * math.exp(self.x)
        return grads
 
 
a, b, c, d = 2, 3, 2, 2
multiply_op1 = multiply()
f1=multiply_op1(a,b)
multiply_op2 = multiply()
f2=multiply_op2(c,d)
add_op = add()
f3=add_op(f1,f2)
exp_op = exponential()
f4=exp_op(f3)
 
print(f4)
 
val1=exp_op.backward(grads=1)
val2=add_op.backward(val1)
val3=multiply_op1.backward(val2[0])
val4=multiply_op2.backward(val2[0])
print(val3)
print(val4)