interpolate — xarray_extras 0.6.1.dev7+g2fd6faf.d20250305 documentation (original) (raw)

• interpolate
• View page source

xarray spline interpolation functions

xarray_extras.interpolate.splev(x_new: object, tck: Dataset, extrapolate: bool | str = True) → DataArray

Evaluate the B-spline generated with splrep().

Parameters:

• x_new – Any DataArray with any number of dims, not necessarily the original interpolation dim. Alternatively, it can be any 1-dimensional array-like; it will be automatically converted to a DataArray on the interpolation dim.
• tck (xarray.Dataset) –
  As returned by splrep(). It can have been:
  • transposed (not recommended, as performance will drop if c is not C-contiguous)
  • sliced, reordered, or (re)chunked, on any dim except the interpolation dim
  • computed from dask to numpy backend
  • round-tripped to disk
• extrapolate –
  True
  Extrapolate the first and last polynomial pieces of b-spline functions active on the base interval
  False
  Return NaNs outside of the base interval
  ’periodic’
  Periodic extrapolation is used
  ’clip’
  Return y[0] and y[-1] outside of the base interval

Returns:

DataArray with all dims of the interpolated array, minus the interpolation dim, plus all dims of x_new

See splrep() for usage example.

xarray_extras.interpolate.splrep(a: DataArray, dim: Hashable, k: int = 3) → Dataset

Calculate the univariate B-spline for an N-dimensional array

Parameters:

• a (xarray.DataArray) – any DataArray
• dim – dimension of a to be interpolated. a.coords[dim] must be strictly monotonic ascending. All int, float (not complex), or datetime dtypes are supported.
• k (int) –
  B-spline order:

  k	interpolation kind
  0	nearest neighbour
  1	linear
  2	quadratic
  3	cubic

Returns:

Dataset with t, c, k (knots, coefficients, order) variables, the same shape and coords as the input, that can be passed to splev().

Example:

x = np.arange(0, 120, 20) x = xarray.DataArray(x, dims=['x'], coords={'x': x}) s = xarray.DataArray(np.linspace(1, 20, 5), dims=['s']) y = np.exp(-x / s) x_new = np.arange(0, 120, 1) tck = splrep(y, 'x') y_new = splev(x_new, tck)

Features

• Interpolate a ND array on any arbitrary dimension
• dask supported on both on the interpolated array and x_new
• Supports ND x_new arrays
• The CPU-heavy interpolator generation (splrep()) is executed only once and then can be applied to multiple x_new (splev())
• memory-efficient
• Can be pickled and used on dask distributed

Limitations

• Chunks are not supported along dim on the interpolated dimension.