Release Notes

Starting with the release of Pastas 0.15 changes to the API will be reported here. The release notes for previous releases up to 0.14 can be found at the GitHub Release page. For full details of all changes check the commit log.

Version 0.15 (31st of July 2020)

Note

This release will introduce backward incompatible changes to Pastas, most notably due to the weighting of the first value of the noise. This will cause the calibrated values to be slightly different but better for most models. It is highly recommended to upgrade to this new version of Pastas.

New Features / Enhancements

  • Model.noise now returns the noise and not the weighted noise. Weights may now be obtained through Model.noise_weights.

  • Private methods are now identified by a leading underscore :issue:`74`.

  • Model.set_parameter method on the Model class is introduced to set the initial, minimum, maximum and vary settings for a parameters in one line.

  • the ps.stats subpackage has been completely restructured. All methods may now also be used as separate methods.

    • ps.stats.diagnostics: perform multiple diagnostic tests at once.

    • ps.stats.stoffer_toloi: Ljung-box test adapted for missing data.

    • ps.stats.plot_diagnostics: stand-alone version of the plot for diagnostic checking

    • ps.stats.plot_acf: convenience method to plot the autocorrelation function.

    • all goodness-of-fit metrics are now available as separate functions e.g., ps.stats.nse. See the API docs for all available methods.

  • A new experimental noise model is added: ArmaModel. This model computes the noise from the residuals according to a autoregressive-moving-average model (ARMA(1,1)). Currently this method is experimental and only applicable to time series with equidistant time steps.

  • The response functions have been standardized to all fit the same formula for the impulse response function, when some parameters are fixed to certain values.

  • new function ps.show_versions is introduced. This function may be used to show the version of package dependencies that are installed.

  • New method ml.get_response_tmax is introduced. This method may be used to obtain the tmax of the response function.

Deprecations

  • ml.set_vary, ml.set_initial, ml.set_pmin, and ml.set_pmax are deprecated and will be removed in a future release . The use of ml.set_parameter method is now recommended.

Backwards incompatible API changes

  • The parameters of the Hantush response function have new names. This will cause problems when loading models using this function to be loaded from .pas-file. No fix is available for this.

  • The first value of the noise series has changes (see :issue:`152` for details), causing changes in the optimal parameter values.

New Example (Notebooks)

  • Notebook on diagnostic checking of Pastas models.

  • Notebook on the new ArmaModel noise model.

  • Notebook on reading Dutch datasets.

  • Notebook on the autocorrelation function with irregular time steps.