Crate tang_ad 

Reverse-mode automatic differentiation.

Tape-based AD for efficient gradient computation when outputs are scalar and inputs are many (the ML/physics optimization case).

Structs

Tape

The AD tape that records the computational graph.

Var

A variable tracked on the AD tape.

Functions

grad

Compute gradient of scalar-valued function via reverse-mode AD.

hessian

Compute Hessian of scalar-valued function via forward-over-forward (Dual<Dual>).

jacobian_fwd

Compute Jacobian via forward-mode (Dual numbers).

jvp

Jacobian-vector product via forward-mode: J v

vjp

Vector-Jacobian product via reverse-mode: v^T J