'Tensorflow autodiff with respect to matrix entry

In the following code, the Jacobian I get is None. Why is that?

a1 = tf.constant(2.0)
a2 = tf.constant(1.5)


with tf.GradientTape(persistent=True) as tape:
    tape.watch(a1)
    tape.watch(a2)
    A = tf.Variable([[a1, 0.0], [0.0, a2]])
    y = tf.linalg.solve(A, x)

dy_da1 = tape.jacobian(y, a1)
dy_da1.numpy()

tensorflow autodiff

Sources

This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.

Source: Stack Overflow

Solution	Source

'Tensorflow autodiff with respect to matrix entry

Sources

Related Questions