2.3 自动求梯度

小白学深度学习

 
 

1. 示例

x = tf.reshape(tf.Variable(range(4), dtype = tf.float32), (4, 1))

with tf.GradientTape() as t:
    t.watch(x)
    y = 2 * tf.matmul(tf.transpose(x), x)

dy_dx = t.gradient(y, x)

 
 

2. 训练模式和预测模式

with GradientTape(persistent=True) as g:
    g.watch(x)
    y = x * x
    z = y * y
    dz_dx = g.gradient(z, x)
    dy_dx = g.gradient(y, x)

dz_dx
dy_dx

 
 

3. 对Python控制流求梯度

def f(a):
    b = a * 2
    while tf.norm(b) < 1000:
        b = b * 2
    if tf.reduce_sum(b) > 0:
        c = b
    else:
        c = b * 100
    return c

a = tf.random.normal((1, 1), dtype = tf.float32)
with GradientTape() as t:
    t.watch(a)
    c = f(a)
t.gradient(c, a) == c/a