evalmodels: Fitting and Comparing Zero-Censored Gaussian Models on...

In RainTyrol: Precipitation Observations and NWP Forecasts from GEFS

Description

The function evalmodels fits distributional trees (disttree), distributional forests (distforest), a prespecified GAMLSS (gamlss), a boosted GAMLSS (gamboostLSS), and an EMOS model (crch) to precipitation data. The results are compared based on CRPS, log-likelihood and RMSE.

Usage

evalmodels(station, train, test, 
           ntree = 100, distfamily = "gaussian",
           tree_minsplit = 50, tree_minbucket = 20, tree_mincrit = 0.95,
           forest_minsplit = 50, forest_minbucket = 20, forest_mincrit = 0,
           forest_mtry = 27,
           gamboost_cvr = FALSE)

Arguments

station	character, name of the selected observation station.
train	numeric, (vector of) years the models should be trained on (available: 1985–2012)
test	numeric, (vector of) years the models should be tested on (available: 1985–2012)
ntree	numeric, number of trees in the distributional forest.
distfamily	character, name of the distribution that should be used, can be either a gaussian or a logistic distribution.
tree_minsplit	numeric, the minimum sum of weights in a node in order to be considered for splitting in the distributional tree.
tree_mincrit	numeric, the value of the test statistic or 1 - p-value that must be exceeded in order to implement a split in the distributional tree.
tree_minbucket	numeric, the minimum sum of weights in a terminal node in the distributional tree.
forest_minsplit	numeric, the minimum sum of weights in a node in order to be considered for splitting in the distributional forest.
forest_minbucket	numeric, the minimum sum of weights in a terminal node in the distributional forest.
forest_mincrit	numeric, the value of the test statistic or 1 - p-value that must be exceeded in order to implement a split in the distributional forest.
forest_mtry	numeric, number of input variables randomly sampled as candidates at each node for random forest like algorithms. The default mtry = Inf means that no random selection takes place.
gamboost_cvr	logical, Should cvrisk be applied to find the optimal value for 'mstop'.

Value

evalmodels returns a list with the following components:

CRPS	CRPS (continuos ranked probability score) of all methods, average over testing data.
LS	Logarithmic score (= log-likelihood) of all methods, average over testing data.
RMSE	Root mean squared error of all methods, average over testing data.

Examples

if(require("crch") &
   require("disttree") &
   require("gamlss") &
   require("gamlss.dist") &
   require("gamlss.cens") &
   require("gamboostLSS") &
   require("mboost") &
   require("partykit") &
   require("scoringRules") &
   require("survival")
) {

evalmodels(station = "Axams", train = 1985:2008, test = 2009:2012, distfamily = "gaussian")

}

RainTyrol documentation built on Jan. 14, 2020, 3:02 a.m.