scipy.spatial.distance.

chebyshev#

scipy.spatial.distance.chebyshev(u, v, w=None)[source]#

Compute the Chebyshev distance.

The Chebyshev distance between real vectors \(u \equiv (u_1, \cdots, u_n)\) and \(v \equiv (v_1, \cdots, v_n)\) is defined as [1]

\[d_\textrm{chebyshev}(u,v) := \max_{1 \le i \le n} |u_i-v_i|\]

If a (non-negative) weight vector \(w \equiv (w_1, \cdots, w_n)\) is supplied, the weighted Chebyshev distance is defined to be the weighted Minkowski distance of infinite order; that is,

\[\begin{split}\begin{align} d_\textrm{chebyshev}(u,v;w) &:= \lim_{p\rightarrow \infty} \left( \sum_{i=1}^n w_i | u_i-v_i |^p \right)^\frac{1}{p} \\ &= \max_{1 \le i \le n} 1_{w_i > 0} | u_i - v_i | \end{align}\end{split}\]
Parameters:
u(…, N) array_like

Input vector.

v(…, N) array_like

Input vector.

w(…, N) array_like, optional

Weight vector. Default is None, which gives all pairs \((u_i, v_i)\) the same weight 1.0.

Returns:
chebyshevfloat or array

The Chebyshev distance between vectors u and v, optionally weighted by w.

Notes

Array API Standard Support

chebyshev has support for Python Array API Standard compatible backends in addition to NumPy. The following combinations of backend and device (or other capability) are supported.

Library

CPU

GPU

NumPy

✅

n/a

CuPy

n/a

✅

PyTorch

✅

✅

JAX

✅

✅

Dask

✅

n/a

See Array API coverage for more information.

References

Examples

>>> from scipy.spatial import distance
>>> distance.chebyshev([1, 0, 0], [0, 1, 0])
1.0
>>> distance.chebyshev([1, 1, 0], [0, 1, 0])
1.0