HSMclass: HSMclass

In AndreasChristianHill/classoptimr: Optimal classification schemes for prediction models with continuous response variable

Description

Identification of optimal classification schemes by Heuritic Search Method (HSM) by Simulated Annealing

Usage

HSMclass(refdata, predictions, nclasses, moveinterval = 10, iterations,
  coolfactor, InitTemp, weight.norefs, weight.classwidth,
  bestever.iteration = 10, progressbar = TRUE, trace = FALSE)

Arguments

refdata	vector containing the values of the continuous response variable used in the prediction model
predictions	predictions for the response values of the prediction model. Note:(refdata and predictions have to correspond to each other)
nclasses	number of classes for which an optimal classification scheme should be computed
moveinterval	controls if classes should be whole numbers (default is 10).
iterations	number of iterations used in heuristic
coolfactor	cooling factor of Simulated Annealing algorithm
InitTemp	intial Temperature of Simulated Annealing algorithm
weight.norefs	weight for maximizing number of reference data pefor each class
weight.classwidth	weight for minimizing the classwidth for each class
bestever.iteration	number of times the heuristic is repeatedly applied. If > 1, the best solution over all runs will be chosen as the optimal solution. Defaults to 10.
progressbar	Shows the progress of the heuristic. Defaults to TRUE.
trace	logical. If TRUE, prints current solution- and penalty-term values to the console.

Value

HSMclass returns an object of class "hsmclass".

An object of class "hsmclass" returns a list of the following components:

best.classbreaks	codevector containing the class break values of the optimal classification scheme
best.classwidth	codevector containing the class width of each class
no.refs.best	codevector containing the number of reference data in each class
BestSolution	value of best solution found by heuristic
Solutions	codevector containing all solution values of heuristic
bestever.iterationmode	codevector containing the class break values of the optimal classification scheme
Temperature	codevector containing the temperature values of the Simulated Annealing algorithm
deltaF	codevector containing the differences between new solution and best solution at the respective iteration of the heuristic
p	codevector containing the p-values (...). Improvement of Sol.Best always yields p = 1
moved.per.iteration	codevector containing the number of classbreaks moved for respective iteration
comp.time	information about the computation time
call	the function call passed to function HSMclass
settings	a list containing the function's inputs: refdata: predictions: nclasses: iterations: moveinterval: coolfactor: InitTemp: weight.norefs: weight.classwidth:

References

Hill, A., Breschan, J., & Mandallaz, D. (2014). Accuracy assessment of timber volume maps using forest inventory data and LiDAR canopy height models. Forests, 5(9), 2253-2275.

Examples

# ------------------------------------------------#
## PERFORM OPTIMIZATION OF CLASSIFICATION SCHEME:

#----
# 1.) Example: optimization for 6 classes:
#----

hsm.1<- HSMclass(refdata = refdata.gr, predictions = predictions.gr,
                 nclasses = 6,iterations = 1000, coolfactor=0.99,
                 InitTemp = 80, weight.norefs = 2, weight.classwidth = 2)
summary(hsm.1)


#----
# 2.) Example: optimization for 6 classes, run heuristic 100 times
#----           and pick best solution over all runs:
 ## Not run: 
 hsm.2<- HSMclass(refdata = refdata.gr, predictions = predictions.gr,
                  nclasses = 6,
                  iterations = 1000, coolfactor=0.99, InitTemp = 80,
                  weight.norefs = 2, weight.classwidth = 2,
                  bestever.iteration = 100)
summary(hsm.2)

## End(Not run)


# ------------------------------------------------#
##  PERFORM ENTIRE ANALYSIS:

# define a set of equidistant intervals to evaluate:
equal.intervals<- seq(100,300,20)

# define corresponding number of classes:
n.classes<- ceiling(max(refdata.gr, predictions.gr)/equal.intervals)

# Chain of analysis:
# --> 1. Identify optimal classification scheme for all given number of classes
# --> 2. Calculate classification accuracy for equidistant class intervals
# --> 3. Calculate classification accuracy for corresponding optimal no. of classes

## Not run: 
acc.equal<- list()
acc.opti<-  list()

lapply(seq_along(n.classes), function(x){

  hsm<- HSMclass(refdata.gr, predictions.gr, nclasses = n.classes[x],
                 iterations = 1000, coolfactor = 0.99, InitTemp = 80,
                 weight.norefs = 2, weight.classwidth = 2)

 acc.equal[[x]]<- classAccuracy(refdata.gr, predictions.gr, equal.int = equal.intervals[x])
 acc.opti[[x]]<- classAccuracy(hsm)

})

## End(Not run)

AndreasChristianHill/classoptimr documentation built on May 29, 2019, 12:23 p.m.