Roadmap
This version
Breaking changes
file reorganization means that any functions that were previously accessed via full path may not be available
qp.[parameterization]now aliases to the parameterization class instead of its create function. However, Ensemble creation has been updated to take the class and not the creation function, so Ensemble functionality should be unbroken. If you want to specify a specific creation method, it takes themethodargument.changed
qp.Ensemble.objdata()andqp.Ensemble.metadata()calls to be properties, to matchqp.Ensemble.ancilqp.Ensemble.metadata['coord'], where ‘coord’ is the coordinates value (i.e. bins or xvals), has been changed from a 2D array to a 1D array in memory. It will still be changed to a 2D array for writing to file.check_inputhas been changed as a parameter to be more descriptive, so it isnormorensure_extentas appropriateoutput of
qp.Ensemblestatistics functions, i.e.pdf, will have their dimensionality dictated by the dimensionality of the given Ensemble and the input value, as inscipy.stats.A single distribution Ensemble will return 1D arrays in the
objdatadictionary in memory. This is true for any sliced Ensembles as well.for parameterizations based on
scipy.statsdistributions, input is automatically reshaped into arrays that are (\(n_{pdf}\), 1) in shape, to ensure that the Ensemble object behaves similarly across all parameterization typespdf_constructor_name(for the quantile parameterization) is now byte-encoded and able to be written to a file
Moving forward
separate out
metricsinto its own packageseparate out
irregular_interpandinterpparameterizations into separate folderspossibly change
qp.Ensemble.npdftoqp.Ensemble.ndist, part of larger change away from calling individual distribution objects PDFsupdate the tests to use
pytestconsistently instead ofunittesthave a metadata translation layer to user datatypes (i.e.
ens.metadata['pdf_name']returns a string not an array of a byte-encoded string)