Struct SvdCotangent 

pub struct SvdCotangent<T: Scalar> {
    pub u: Option<Tensor<T>>,
    pub s: Option<Tensor<T>>,
    pub vt: Option<Tensor<T>>,
}

Cotangent (adjoint) for SVD outputs.

Each field is Option because the user may not need gradients for all outputs (e.g., only s for singular value optimization).

Examples

use tenferro_linalg::SvdCotangent;
use tenferro_tensor::{Tensor, MemoryOrder};
use tenferro_device::LogicalMemorySpace;

let col = MemoryOrder::ColumnMajor;
let mem = LogicalMemorySpace::MainMemory;

// Only need gradient through singular values
let cotangent = SvdCotangent {
    u: None,
    s: Some(Tensor::<f64>::ones(&[3], mem, col)),
    vt: None,
};

Fields

u: Option<Tensor<T>>

Cotangent for U. Shape must match SvdResult::u.

s: Option<Tensor<T>>

Cotangent for S. Shape must match SvdResult::s.

vt: Option<Tensor<T>>

Cotangent for Vt. Shape must match SvdResult::vt.