Exact floating-point equality is rarely what you want after arithmetic. allclose() checks every element pair with the formula:
Available on Vector, VectorView, VectorSpan, Tensor, TensorView, and TensorSpan.
For tensors, allclose is provided by the AllCloseOps trait — import it if calling on a TensorRef implementor.
Shape mismatch returns false.
The scalar helper is_close is re-exported at crate root.
use numkong::{is_close, Vector, Tensor};
// Scalar check
assert!(is_close(1.0, 1.0 + 1e-8, 1e-6, 0.0));
// Vector tolerance check
let a = Vector::<f32>::full(3, 1.0).unwrap();
let b = Vector::<f32>::full(3, 1.0 + 1e-7).unwrap();
assert!(a.allclose(&b, 1e-6, 0.0));
// Tensor tolerance check
let ta = Tensor::<f32>::full(&[2, 3], 1.0).unwrap();
let tb = Tensor::<f32>::full(&[2, 3], 1.0 + 1e-7).unwrap();
assert!(ta.allclose(&tb, 1e-6, 0.0));AllCloseOpsExtension trait: tolerance-based equality for any TensorRef implementor.1pub trait AllCloseOps<Scalar: FloatConvertible, const MAX_RANK: usize>: TensorRef<Scalar, MAX_RANK>
where
Scalar::DimScalar: NumberLike,Uses the formula |a − b| ≤ atol + rtol × |b| per element. Returns false if shapes differ.
impl<C, Scalar: FloatConvertible, const R: usize> AllCloseOps<Scalar, R> for C where C: TensorRef<Scalar, R>, Scalar::DimScalar: NumberLikeallcloseReturns true iff both tensors share a shape and every element pair is within tolerance.fn allclose(&self, other: &impl TensorRef<Scalar, MAX_RANK> + ?Sized, atol: f64, rtol: f64) -> boolis_closeCheck whether two values are within tolerance: |a − b| ≤ atol + rtol × |b|.pub fn is_close(a: f64, b: f64, atol: f64, rtol: f64) -> bool