Changelog
Source:NEWS.md
PEcAn.uncertainty 1.9.0.9000
-
generate_joint_ensemble_design(sobol = TRUE)now errors when an input is sampled by"looping"or parameters by"halton"or"sobol". These split into dependentX1andX2halves and gave wrong Sobol indices (#4134). - In
run.sensitivity.analysis(), loadsamplesonce before the variables loop, avoiding repeated disk reads (#3859). -
get.ensemble.samples()now again returns a list of dataframes as it did before v1.9.0. The returned list also now has attributesampled.indices, giving the position of each sampled value in its input set (#4127, @Devansh-56) -
get.results()reads sensitivity output only for the PFTs that have runs insa.run.ids, so PFTs not at a site no longer give a warning per trait and quantile. -
read.sa.output()now reads only the runs insa.run.ids, so each site of a multisite sensitivity analysis gets its own results. Previously every site read the last site’s runs from the sharedruns_manifest.csv. -
run.sensitivity.analysis()skips traits with no spread in their SA samples, with a warning. Previously one such trait stopped the analysis for every PFT. -
runs_manifest.csvnow has a columnens_numto distinguish ensemble members without reparsing run IDs. -
run.sensitivity.analysis()no longer converts traits in Celsius (e.g.psnTOpt) to K before evaluating their splines, which are fit in Celsius. Their sensitivity, elasticity and variance were wrong in earlier versions, and so were the partial variances of the other traits in the PFT. CV and elasticity are still computed in K. - Multiple bugfixes in
input.ens.gen()handling of parent ids (#3783):- No longer skips inputs that have a parent but no sampling method.
- Argument
parent_idsnow accepts integer vectors even if not wrapped in a list. - New argument
bad_parent_actionchooses how to handle parent ids that do not match an input path: default “error” to stop with an error, “resample” to replace invalid indices from the input paths that do exist.
- run.ensemble.analysis() now respects
settings$modeloutdirrather than assuming anout/folder insidesettings$outdir(@Akash-paluvai, #3722). - Added
generate_OAT_SA_design()for creating input design matrices for sensitivity analysis. This function ensures non-parameter inputs (met, IC, soil) are held constant, which is required for valid variance decomposition in one-at-a-time (OAT) sensitivity analysis (#3729) -
get.parameter.samples()now consistently accepts PFTs with nooutdirspecified. These previously failed when no database connection was available. - Improved row alignment and text scaling in
plot_variance_decomposition(). -
plot_variance_decomposition()gains argumentorder_by, sorting highest variance to the top by default. To reproduce old behavior, useorder_by = "none".
PEcAn.uncertainty 1.9.0
License change
PEcAn.uncertainty is now distributed under the BSD 3-clause license instead of the NCSA Open Source license.
Added
- Added the shared
input_designmatrix, generated via new funtiongenerate_joint_ensemble_design(), keeps parameter draws and sampled inputs aligned acrossrun.write.configs(),write.ensemble.configs(), andwrite.sa.configs(). (@blesson-07, #3535, #3677). - New function
compute_sobol_indicesfor use in global sensitivity analyses (@blesson-07, #3612).
Changed
- Breaking:
write.ensemble.configsgains new mandatory argumentsinput_designandensemble.size, and removes argumentsamples. Note thatinput_designandensemble.sizeare added to the beginning of the argument list, so calls passingdefaultsunnamed as the first arg will break (#3612, #3634). - Breaking:
input.ens.gengains new mandatory argumentensemble_size, added between the existingsettingsandinputargs. - Plotting sensitivity now makes less noise in the console and once again produces a one-page PDF as intended (#3560).
-
write.ensemble.configsandwrite.sa.configsnow generate an ensemble id if one is not provided in a DB-free run (#3654). Runs with DB continue to always generate a new id. Note that multi-site runs with no id provided will now get a separate ensemble ID (and thus generate separate analyses) for each site. - Documented that
runModule.run.sensitivity.analysisdoes not yet work with multisite settings. This will be fixed in a future release. - The default sampling method of
get.ensemble.sampleshas changed from “uniform” to “random” (#3535).
PEcAn.uncertainty 1.8.1
-
write.ensemble.configsnow respects argumentwrite.to.dbwhen settingsbetydatabasewrite is set, it still overrideswrite.to.db). - Roxygen fixes
PEcAn.uncertainty 1.8.0
- Added an optional
pftsargument torun.sensitivity.analysis()so that sensitivity analysis and variance decomposition can be run on a subset of PFTs defined insettingsif desired (#3155).