Quick search


version 0.6

Framework for fitting functions to data with SciPy

People: Ludwig Schwardt


Fitting SciKit

A framework for fitting functions to data with SciPy which unifies the various available interpolation methods and provides a common interface to them based on the following simple methods:

  • Fitter.__init__(p): set parameters of interpolation function, e.g. polynomial degree
  •, y): fit given input-output data
  • Fitter.__call__(x) or Fitter.eval(x): evaluate function on new input data

Each interpolation routine falls in one of two categories: scatter fitting or grid fitting. They share the same interface, only differing in the definition of input data x.

Scatter-fitters operate on unstructured scattered input data (i.e. not on a grid). The input data consists of a sequence of x coordinates and a sequence of corresponding y data, where the order of the x coordinates does not matter and their location can be arbitrary. The x coordinates can have an arbritrary dimension (although most classes are specialised for 1-D or 2-D data). If the dimension is bigger than 1, the coordinates are provided as an array of column vectors. These fitters have ScatterFit as base class.

Grid-fitters operate on input data that lie on a grid. The input data consists of a sequence of x-axis tick sequences and the corresponding array of y data. These fitters have GridFit as base class.

The module is organised as follows:

Scatter fitters

  • ScatterFit: Abstract base class for scatter fitters
  • LinearLeastSquaresFit: Fit linear regression model to data using SVD
  • Polynomial1DFit: Fit polynomial to 1-D data
  • Polynomial2DFit: Fit polynomial to 2-D data
  • PiecewisePolynomial1DFit: Fit piecewise polynomial to 1-D data
  • Independent1DFit: Interpolate N-dimensional matrix along given axis
  • Delaunay2DScatterFit: Interpolate scalar function of 2-D data, based on Delaunay triangulation and cubic / linear interpolation
  • NonLinearLeastSquaresFit: Fit a generic function to data, based on non-linear least squares optimisation
  • GaussianFit: Fit Gaussian curve to multi-dimensional data
  • Spline1DFit: Fit a B-spline to 1-D data
  • Spline2DScatterFit: Fit a B-spline to scattered 2-D data
  • RbfScatterFit: Do radial basis function (RBF) interpolation

Grid fitters

  • GridFit: Abstract base class for grid fitters
  • Spline2DGridFit: Fit a B-spline to 2-D data on a rectangular grid

Helper functions

  • squash: Flatten array, but not necessarily all the way to a 1-D array
  • unsquash: Restore an array that was reshaped by squash
  • sort_grid: Ensure that the coordinates of a rectangular grid are in ascending order
  • desort_grid: Undo the effect of sort_grid
  • vectorize_fit_func: Factory that creates vectorised version of function to be fitted to data
  • randomise: Randomise fitted function parameters by resampling residuals


Ludwig Schwardt <ludwig at>


0.6 (2016-12-05)

  • Fix pip installation, clean up setup procedure, flake8 and add README
  • PiecewisePolynomial1DFit updated to work with scipy 0.18.0
  • Delaunay2DScatterFit now based on scipy.interpolate.griddata, which is orders of magnitude faster, more robust and smoother. Its default interpolation changed from 'nn' (natural neighbours - no longer available) to 'cubic'.
  • Delaunay2DGridFit removed as there is no equivalent anymore

0.5.1 (2012-10-29)

  • Use proper name for np.linalg.LinAlgError

0.5 (2011-09-26)

  • Initial release of scikits.fitting



You can download the latest distribution from PyPI here:

Using pip

You can install scikits.fitting for yourself from the terminal by running:

pip install --user scikits.fitting

If you want to install it for all users on your machine, do:

pip install scikits.fitting
On Linux, do sudo pip install scikits.fitting.

If you don't yet have the pip tool, you can get it following these instructions.

This package was discovered in PyPI.