Changelog
Source:NEWS.md
harpPoint v0.2.1
HOTFIX released on 2nd February 2024
Fixes a bug that caused errors in
check_obs_against_fcst()
for deterministic forecastsFixes bug that meant
lag_forecast()
andshift_forecast()
did not accept the harp >= v0.2.0 column naming convention
harpPoint v0.2.0
This is a major update. Most of the changes are internal, but there are some changes that may lead to problems with downstream scripts that are listed below along with some selected new features.
Breaking changes
Default
groupings
argument changed from"leadtime"
to"lead_time"
for consistency with changes in {harpIO}. Scripts that use verification functions with"leadtime"
,"validdate"
and"fcdate"
as values togroupings
should be changed to"lead_time"
,"valid_dttm"
and"fcst_dttm"
respectively.Verification outputs now use the column name
fcst_model
instead ofmname
so that there is consistency throughout harp.Attributes for verification outputs have changed to include all forecast dates, all stations and all groupings used in the verification. Any scripts that make use of these attributes should be updated to reflect the new attributes.
scale_point_forecast()
andscale_point_obs()
are deprecated.scale_param()
should be used instead.gather_members()
andspread_members()
are deprecated.pivot_members()
should be used instead.first_validdate()
andlast_validdate()
are deprecated.unique_valid_dttm()
should be used instead.pull_stations()
is deprecated.unique_stations()
should be used instead.bootstrap_score()
,pooled_bootstrap_score()
andbind_bootstrap_score()
are defunct.bootstrap_verify()
andbind_point_verif()
should be used instead.
Selected new features
Verification functions have gained new progress bars and are generally less verbose in what they are doing, restricting messages to only progress with computing scores for different verification groups.
New verification score hexbin. This gives a data frame of what is essentially a heat map of forecast - observation value pairs.
New class, attributes and print method for verification function outputs.
jitter_fcst()
now accepts vectorized functions so should be a lot faster.