First CRAN release.
params,
using Earth Engine’s argument names, for example
params = list(gbt = list(shrinkage = 0.01)). This replaces
the separate arguments n_trees, shrinkage,
svm_cost, knn_k and the rest, which applied to
several models at once and could silently override values equal to their
defaults. Misspelt settings are now an error.rf 500, gbt 150) and
knn, which uses 15 neighbours because Earth Engine’s single
neighbour gives a two-value map. The SVM is now Earth Engine’s default
linear classifier, and boosted trees use Earth Engine’s learning rate,
which gives better-calibrated probabilities.generate_pseudo_absences() now reads pseudo-absences
from the same years as the presences, in the same proportions, instead
of the latest embedding year. Pass aoi_year to place them
all in one year.balance_trees is gone (use
bg_ratio, default 1, or NULL for every
absence), as are persist_classifier, async and
options from evaluate_models(), which now
requires predict_coords with a year column.
generate_map() maps the records’ most common year by
default.sdm_gee_status() is now gee_tasks() and
sdm_clean_assets() is gee_clean_assets().
sdm_verbose() and the exported calculate_cbi()
are gone: use suppressMessages(), and the cbi
from calculate_classifier_metrics().generate_map() computes the ensemble on Earth Engine
and accepts aoi = "bbox".reticulate::py_require(), so reticulate installs it when
needed; AlphaSDM no longer installs Python itself and no longer depends
on ‘rgee’.tools::R_user_dir("AlphaSDM", "config").
clear_gee_credentials() removes it and asks before signing
out of Earth Engine.gee_clean_assets().vignette("AlphaSDM"), maps
saguaro around Tucson from GBIF records.