Releases: ropensci/waywiser
waywiser 0.6.0
-
Metric functions now return
NA
in all cases where they previously returnedNaN
.
This improves cross-platform consistency; in particular, MacOS often returned
NA
when every other platform would returnNaN
. (#63) -
ww_multi_scale()
now handles classification and class probability metrics better
when called with raster arguments (either todata
or totruth
andestimate
):- When called with classification metrics,
ww_multi_scale()
will now convert
truth
andestimate
to factors before passing them to the metric set.
Thanks to @Nowosad for the report in #60 (#61). - When called with class probability metrics,
ww_multi_scale()
will convert
truth
to a factor and will passestimate
as an unnamed argument. (#62) - When called with a mix of class and probability metrics,
ww_multi_scale()
will error. (#62)
- When called with classification metrics,
waywiser 0.5.1
-
ww_multi_scale()
now warns if you providecrs
as an argument tosf::st_make_grid()
via...
. Grids created by this function will always take their CRS fromdata
. -
ww_multi_scale()
now throws an error if you pass arguments via...
while also providing a list of grids (because those arguments would be ignored). -
ww_multi_scale()
is now faster whendata
is an sf object, particularly when grids are created by passing arguments tosf::st_make_grid()
(rather than passing grids viagrids
).
waywiser 0.5.0
-
ww_multi_scale()
did not correctly handle grid cellsizes with units. Units (set using theunits
package) are now respected. -
ww_multi_scale()
using sf data always returned "1" for truth and estimate counts. This was because counts were calculated post-aggregation by mistake. Calculation order has been fixed, and these counts should now be correct.
waywiser 0.4.2
- This is a patch release to keep up with additional breaking changes in the vip package. There are no user-facing differences in this version.
waywiser 0.4.1
- This is a patch release to keep up with breaking changes in the vip package. There are no user-facing differences in this version.
waywiser 0.4.0
Breaking Changes
-
ww_build_neighbors()
(and by extent, every spatial dependence metric) will
no longer calculate neighbors for non-point or non-polygon geometries. -
Functions dealing with local Moran's I and related p-values (both data frame
and vector variants) now return unnamed vectors.
New Features
-
Added a new method to support passing a
SpatRaster
todata
in
ww_multi_scale()
, withtruth
andestimate
being indices used to subset
data
. This is a bit faster than passingSpatRaster
objects totruth
and
estimate
, as extraction is only done once per grid rather than twice, but
does not easily support passing R functions toaggregation_function
. -
Metric functions now have better error messages including the name of the
function the user called that errored, not the internal function that
errored. Huge thanks to @EmilHvitfeldt for their PR (#40). -
Data frame metric functions now guarantee that
.estimate
will be an unnamed
vector.
Bug Fixes
-
The
sf
method forww_multi_scale()
is now much faster (and more memory
efficient). -
Fixed the warning when
ww_area_of_applicability()
calculates an AOA
threshold of 0. It now includes "Did you accidentally pass the same data as
testing and training?" as a bullet.
waywiser 0.3.0
Breaking Changes
-
Removed combination functions --
ww_global_geary
,ww_global_moran
,
ww_local_geary
,ww_local_moran
,ww_local_getis_ord
. Usemetric_set()
to combine functions instead. -
Renamed
ww_local_getis_ord_pvalue_vec()
and variants to
ww_local_getis_ord_g_pvalue_vec()
; this change allows internal functions to
work properly, and makes it easier for the output to indicate if the p-value
is associated with a g or g* value. -
Yardstick metrics will no longer include geometry columns in their returns.
-
The
na_action
argument toww_area_of_applicability()
has been replaced by
na_rm
, with a default value ofFALSE
. -
na_rm
is nowTRUE
by default for non-spatial-autocorrelation functions.
NA values will cause spatial-autocorrelation functions to fail with an error.
New Features
-
Added functions (primarily
ww_multi_scale()
) and a vignette for multi-scale
assessment of model predictions. -
Added functions to calculate metrics from Ji and Gallo (2006) and Willmott
(1981, 1982, 2012):ww_agreement_coefficient()
,
ww_systematic_agreement_coefficient()
,
ww_unsystematic_agreement_coefficient()
,ww_unsystematic_mpd(),
,
ww_systematic_mpd()
,ww_unsystematic_rmpd()
,ww_systematic_rmpd()
,
ww_willmott_d()
,ww_willmott_dr()
,ww_willmott_d1()
,
ww_systematic_mse()
,ww_unsystematic_mse()
,ww_systematic_rmse()
,
ww_unsystematic_rmse()
. Note thatww_willmott_dr()
uses the version
from Willmott (2012); other implementations (sometimes called "d1r") seem to
use an unbounded variant that I haven't found a reference to support. -
Changed a call in
ww_area_of_applicability()
to use FNN for nearest
neighbors, rather than fields. This sped up prediction by a lot.
Bug Fixes
-
Rewrote README and moved the old content to a new vignette on assessing the
spatial dependency in model residuals. -
Added a dependency on FNN.
-
Minimum version for dplyr has been bumped to 1.1.0
waywiser 0.2.0
- Added functions for calculating the area of applicability of a model.
waywiser 0.1.0
-
Added functions for automatically constructing
nb
andlistw
objects -
Added functions for global and local Geary's C values.
-
Added functions for local Getis-Ord G and G* values.
-
Added functions for global and local Moran's I values.
-
Added a
NEWS.md
file to track changes to the package.