jax.numpy.interp#

jax.numpy.interp(x, xp, fp, left=None, right=None, period=None)[source]#

One-dimensional linear interpolation.

JAX implementation of numpy.interp().

Parameters:

  • x (ArrayLike) – N-dimensional array of x coordinates at which to evaluate the interpolation.

  • xp (ArrayLike) – one-dimensional sorted array of points to be interpolated.

  • fp (ArrayLike) – array of shape xp.shape containing the function values associated with xp.

  • left (ArrayLike | str | None) – specify how to handle points x < xp[0]. Default is to return fp[0]. If left is a scalar value, it will return this value. if left is the string "extrapolate", then the value will be determined by linear extrapolation. left is ignored if period is specified.

  • right (ArrayLike | str | None) – specify how to handle points x > xp[-1]. Default is to return fp[-1]. If right is a scalar value, it will return this value. if right is the string "extrapolate", then the value will be determined by linear extrapolation. right is ignored if period is specified.

  • period (ArrayLike | None) – optionally specify the period for the x coordinates, for e.g. interpolation in angular space.

Returns:

an array of shape x.shape containing the interpolated function at values x.

Return type:

Array

Examples

>>> xp = jnp.arange(10)
>>> fp = 2 * xp
>>> x = jnp.array([0.5, 2.0, 3.5])
>>> interp(x, xp, fp)
Array([1., 4., 7.], dtype=float32)

Unless otherwise specified, extrapolation will be constant:

>>> x = jnp.array([-10., 10.])
>>> interp(x, xp, fp)
Array([ 0., 18.], dtype=float32)

Use "extrapolate" mode for linear extrapolation:

>>> interp(x, xp, fp, left='extrapolate', right='extrapolate')
Array([-20.,  20.], dtype=float32)

For periodic interpolation, specify the period:

>>> xp = jnp.array([0, jnp.pi / 2, jnp.pi, 3 * jnp.pi / 2])
>>> fp = jnp.sin(xp)
>>> x = 2 * jnp.pi  # note: not in input array
>>> jnp.interp(x, xp, fp, period=2 * jnp.pi)
Array(0., dtype=float32)