Module ndarray_support 

Available on crate feature ndarray only.

ndarray adapters for echidna’s bytecode tape AD.

Thin wrappers accepting Array1<F> and returning Array1<F> / Array2<F>.

All functions here accept non-contiguous arrays (slices, transposed views, stepped views). Input data is copied element-wise via iter().copied() before being passed to the tape; the previous .as_slice().unwrap() path panicked on any non-C-contiguous layout, which was inconsistent with the faer/nalgebra adapters.

Functions

grad_ndarray

Record a function and compute its gradient, returning an Array1.

grad_ndarray_val

Record a function, compute value and gradient, returning (value, Array1).

hessian_ndarray

Record and compute the Hessian, returning (value, gradient, hessian).

hvp_ndarray

Compute the Hessian-vector product, returning (gradient, hvp) as Array1.

jacobian_ndarray

Compute the Jacobian of a multi-output function, returning Array2<F>.

sparse_hessian_ndarray

Compute the sparse Hessian, returning (value, gradient, pattern, values).

sparse_jacobian_ndarray

Compute the sparse Jacobian, returning (outputs, pattern, values).

tape_gradient_ndarray

Evaluate gradient on a pre-recorded tape, accepting and returning ndarray types.

tape_hessian_ndarray

Evaluate Hessian on a pre-recorded tape, accepting and returning ndarray types.

tape_hvp_ndarray

Compute the Hessian-vector product on a pre-recorded tape, returning (gradient, hvp).

tape_sparse_hessian_ndarray

Compute the sparse Hessian on a pre-recorded tape.