Wraps tape$gradient() and returns tf$zeros_like(x) on error or
when the result is a Python None.
Arguments
- tape
A tf$GradientTape object.
- y
The tensor to differentiate.
- x
The variable with respect to which to differentiate.
Value
A TensorFlow tensor (gradient or zeros).
