Currently, the gradients of the transformation function are numerically approximated. Providing an option to pass a gradient function may speed up the computation. The main drawback is that users would have to implement a jacobian matrix because the transformation function takes a vector as input and returns a vector.
For the time being I will leave this issue open until there is a practical use case where numerical approximations are too slow.
Currently, the gradients of the transformation function are numerically approximated. Providing an option to pass a gradient function may speed up the computation. The main drawback is that users would have to implement a jacobian matrix because the transformation function takes a vector as input and returns a vector. For the time being I will leave this issue open until there is a practical use case where numerical approximations are too slow.