What’s New

Upcoming release

Fix bug with last numpy and integer management :issue:`42`. (:pr:`43`) by G. Maze

Improved documentation
Improved unit testing
Bug fix:
- Fix a bug in the preprocessing step using dask_ml bakend that would cause an error for data already in dask arrays

Warning

The API has changed, break backward compatibility.

Enhancements:
- Multiple-features classification
- ND-Array classification (so that you can classify directly profiles from gridded products, eg: latitude/longitude/time grid, and not only a collection of profiles already in 2D array)
- pyXpcm methods can be accessed through the xarray.Dataset accessor namespace pyxpcm
- Allow to choose statistic backends (sklearn, dask_ml or user-defined)
- Save/load PCM to/from netcdf files
pyXpcm now consumes xarray/dask objects all along, not only on the user front-end. This add a small overhead with small dataset but allows for PCM to handle large and more complex datasets.