Contributing

If you have an idea or would like to correct something, please submit an issue, or a pull request.

If you have questions or are not sure if something is a bug, feel free to submit an issue, or ask a question tagged scikit-guess on Stack Overflow.

This scikit is still in early stages, so please provide all the criticism and advice you can. Any support at all is welcome. In particular, the following areas are under construction:

If you are really interested in going down this path, please read on:

Project Structure

Each fitting algorithm resides in its own module. All the functions get imported into the base skg namespace. Each module should contain a function called model that applies the fitting parameters to a given set of x-values, either raveled or along a particular axis (assuming the function is 1D). Multiple algorithms that fit to the same model can live in the same module. There is no standard yet for resolving naming conflicts in such cases.

Testing

Testing is done using the pytest framework. A test module for every main module exists in the skg.tests package.

Tests for new modules are generated in a semi-automated manner (still WIP). All the modules containing a fitting function and a model will be tested against randomly generated inputs, and checked for speed and quality. The quality of each algorithm will be assessed based on these tests. Quality has three categories: speed, accuracy and usefulness.

  • Speed is a benchmark against scipy.optimize.curve_fit. An algorithm that is slower than a non-linear optimizer starting with default parameters is not deemed very useful.
  • Accuracy is checked by making sure that the fit is within reasonable bounds of the values computed by scipy.optimize.curve_fit. Reasonableness is a function of the analytically derived partial derivatives of the model with respect to the parameters.
  • Usefulness is a measure of how many iterations scipy.optimize.curve_fit saves by using the algorithm as an initial guess. Another informal metric is the combined runtime of the algorithm and curve_fit vs. the runtime of just curve_fit with default parameters. If the latter exceeds the former, that’s a win.

See the Testing SKG page for information on how to run and modify the tests.