This notebook compares 4 versions of the percentile_doy code.
xclim_023, the old version using groupby+nanpercentiles.
xclim_024, first version of Ouranosinc/xclim#638, using MultiIndex+partial-nanpercentiles.
xclim_dev, using MultiIndex+apply_ufunc (reusing the _calc_perc of xclim.ensembles)
xclim_dev2,using groupby+apply_ufunc.
Ver 2 and 3 are limited to daily datasets.
Has can be seen in the notebook, V2 and V3 are the fastest. V4 suffers from the use of groupby, but that allows it to have no limitations on the sampling frequency.
V3 and V4 are lazy, whereas the 2 others aren't.
This notebook compares 4 versions of the
percentile_doy
code.xclim_023
, the old version usinggroupby+nanpercentiles
.xclim_024
, first version of Ouranosinc/xclim#638, usingMultiIndex+partial-nanpercentiles
.xclim_dev
, usingMultiIndex+apply_ufunc
(reusing the_calc_perc
of xclim.ensembles)xclim_dev2
,usinggroupby+apply_ufunc
.Ver 2 and 3 are limited to daily datasets.
Has can be seen in the notebook, V2 and V3 are the fastest. V4 suffers from the use of
groupby
, but that allows it to have no limitations on the sampling frequency. V3 and V4 are lazy, whereas the 2 others aren't.