[traintest]
#dataset = /home/dbv/bilder/15Scenes/imagesScaled/
dataset = /home/luetz/data/images/15Scenes/imagesScaled/
classselection_train = "*"
classselection_test = "*"
examples_train = random * 100
examples_test = random * 50

[cache]
#root = "/home/rodner/3rdparty/imagenetBOF/niceFeatures/"
root = "/home/luetz/data/feature-storage/15Scenes/niceFeatures/"

[GP_IL]
trainExPerClass = 10
num_runs = 10
do_classification = true
incrementalAddSize = 1
nrOfIncrements = 50

[main]
# extension of all files in the cache
ext = ".feat"

[GPHIKClassifier]
noise =  0.01
parameter_lower_bound = 0.5
parameter_upper_bound = 2.0 
#--define the uncertainty prediction scheme--
# standatd predictive variance
#uncertaintyPrediction = pred_variance
# use the heuristic as proposed by Kapoor et al.
uncertaintyPrediction = heuristic
# no classification uncertainty at all?
#uncertaintyPrediction = none

#--define the computation scheme for the predictive variance, if needed--
#if we do not need any predictive variance for this experiment
#varianceApproximation = none
# predictive variance approximation useful for sparse features - really fast
varianceApproximation = approximate_rough 
# predictive variance approximation with eigenvectors (finer)
#varianceApproximation = approximate_fine
#nrOfEigenvaluesToConsiderForVarApprox = 2
#exact computation of predictive variance
#varianceApproximation = exact

#--define the optimization method--
#optimization_method = none
optimization_method = downhillsimplex

#--stuff for the IterativeLinearSolver--
#ils_verbose = true