//! Automatic differentiation.
//!
//! * [`Scalar`] abstracts the real number type, so every kernel runs on `f64` and on [`Dual`] numbers.
//! * [`Tape`] records operations on real and complex matrices and differentiates a real scalar result in reverse,
//! like PyTorch's autograd.
//! * [`Tape::batch_mean`] evaluates independent sub-graphs in parallel and averages them.
//!
//! Exact Hessian-vector products come from running a tape over [`Dual`] numbers: seeding the inputs with a direction
//! `v`, the derivative part of the gradient is `H·v`.
pub use Dual;
pub use ;
pub use ;
pub use ;