desc.derivatives.AutoDiffDerivative.compute_vjp

classmethod AutoDiffDerivative.compute_vjp(fun, argnum, v, *args, **kwargs)Source

Compute v.T * df/dx.

Parameters:

• fun (callable) – function to differentiate

• argnum (int or tuple) – arguments to differentiate with respect to

• v (array-like or tuple of array-like) – tangent vectors. Should be one for each output of fun.

• args (tuple) – arguments passed to fun

• kwargs (dict) – keyword arguments passed to fun

Returns:

vjp (array-like) – Vector v times Jacobian, summed over different argnums