Package net.simplace.sim.control.selectors

Class WeightedLeastDifferenceSelector

java.lang.Object
net.simplace.sim.control.FWSimulationControlContainer
net.simplace.sim.control.FWSimulationSelector
net.simplace.sim.control.selectors.WeightedLeastDifferenceSelector
All Implemented Interfaces:
FWSimFieldContainer
Direct Known Subclasses:
LeastDifferenceSelector
public class WeightedLeastDifferenceSelector extends FWSimulationSelector
Selects the simulation with the best error/deviation for multiple target variables. The general error is the weighted sum of the errors. WIKI_START == Configuration == User has to specify - the frequency at which values should be compared - the calculation method cMode (Bias,MRE,MAE,RMAE,MSE,RMSE,index,rc etc.) - the number of target variables cNumberOfVariables (up to 20) - for each target variable a pair of SimVariables cObserved{i}/cSimulated{i} - for each target variable the weight for the error cWeight{i} - {i} ranges from 1 to cNumberOfVariables == Outputs == - the selected simulation id as a one size CHARARRAY `selectedsimulations` - the best `ErrorValue` achieved as DOUBLE - all simulations weighted sum of errors `ErrorValues` as DOUBLEARRAY - for each target variable all simulations error `ErrorValuesVariable{i}` as DOUBLEARRAY == Description == For a simulation the calculated values of different variables \(i = 1, ... cNumberOfVariables\) are compared to reference (measured) values (e.g. Yield, LAI, Biomass). The selector can handle maximum 20 different variables. Each variable exists pairwise in the simulation as observed\simulated. Observed variables normally come from a resource (datafile), where simulated ones are calculated. (Notice: The selector works also, if both variables come from resources or if both are simulated.) For each of these variables \(n\) samples are produced by the simulation (\(n\) is determined by the simulated time span and the selectors frequence information). Let \(\mathcal{E}: \mathbb{R}^n \times \mathbb{R}^n\ \rightarrow \mathbb{R}\) be the selected calculation method - e.g. the bias \(E(o,s)=\sum_{j=1}^n o_j - s_j\) The weighted difference is then WIKI_END \[ E = \frac{1}{\omega}\sum_{i=1}^{cNumberOfVariables} cWeight_i \mathcal{E}(cObserved_i,cSimulated_i) \] where \(\omega\) is the sum of weights: \[ \omega = \sum_{i=1}^{cNumberOfVariables} cWeight_i \] WIKI_START If \(k= 1, ... l\) simulations are performed within a project, then the selector calculates all weighted errors \(E_k\) and selects the simulation with the best error. (The best error depends on the calculation method for RMSE it's the one closest to 0, for agreement index closest to 1 etc.) The output `NuberOfValues` counts the number of simulations used for comparison. Normally it's equal to \(l\), but it can be lower because simulations that produce no usable result (i.e. \(E_k\) could not be calculated) are skipped. == Example == {{{ | <selector id="Selector" class="net.simplace.sim.control.selectors.WeightedLeastDifferenceSelector" | frequence="COMPLEX" rule="${observations.HasData}" > | RRMSE | 1 | | | | 0.8 | | | | 0.2 | }}} WIKI_END WIKI_START == Error methods == - `Bias` - mean error - `SignedBias` - mean error, selects the smallest signed bias - `MRE` - mean relative error - `MAE` - mean absolute error - `RMAE` - relative mean absolute error - `MSE` - mean squared error - `RMSE` - root mean squared error - `RRMSE` - relative root mean squared error - `SMAPE` - symmetric mean absolute percentage error - `EF` - model efficiency - `r` - Correlation Coefficient - `rc` - Concordance Correlation Coefficient - `index` - Agreement Index - `NONE` - Void Error Calculator WIKI_END
Author:
Gunther Krauss
See Also: