##################################### ###### SYSTEM PARAMETERS ####### #write out confusion matrix or not classifier.outputconfmatrix=true #remove the arff files generated for input to weka learners classifier.removearff=false #save the model object in file for later use classifier.savemodel=false #write the original values in the forecast file as well classifier.writeoutrealvalues=false #the input files are in arff format (if false C4.5 format required) classifier.arffinput=false #maximum number of different values for a discrete feature (set to 0 to have only numeric features) features.discretevalues.max=0 #class labels listed here (comma separated) will be included in the arff header in any case #(as default, classvalues found in train or test instances are added, use this to list further labels if desperately needed) #note that the wrapper handles only labels that can be converted to numeric values (example: 0,1,2,3 - predefined labels) features.classvalue.predefinedvalues=1,2 ################################################## ######## K NEAREST NEIGHBOR PARAMETERS ########## #The number of neighbours to use. knn.neighbors=1 #Whether hold-one-out cross-validation will be used to select the best k value. knn.crossvalidate=false #Gets the distance weighting method used. # none - no weighting used # div - weight by 1/distance # diff - weight by 1-distance knn.distanceweighting=none #The nearest neighbour search algorithm to use (Default: LinearNN). # kdtree, linearnn knn.nn.searchalgorithm=linearnn #Gets the maximum number of instances allowed in the training pool. #The addition of new instances above this value will result in old instances being removed. #A value of 0 signifies no limit to the number of training instances. knn.windowsize=0 #The max number of instances in a leaf. kdtree.maxinstinleaf=40 #The minimum relative width of the box. #A node is only made a leaf if the width of the split dimension of the instances in a node #normalized over the width of the split dimension of all the instances #is less than or equal to this minimum relative width. kdtree.minboxrelwidth=0.01 #Whether if the widths of the KDTree node should be normalized by the width of the universe or not. #Where, width of the node is the range of the split attribute based on the instances in that node, #and width of the universe is the range of the split attribute based on all the instances (default: false). kdtree.normalizenodewidth=false #Whether to skip identical instances (with distance 0 to the target) linearnn.skipidentical=false ########################################## ######## LIBSVM PARAMETERS ########## #The type of SVM to use #c-svc, nu-svc, oneclass-svm, eps-svr, nu-svr libsvm.svmtype=nu-svc #The cache size in MB libsvm.cachesize=40 #The coefficient to use. libsvm.coef0=0.0 #The cost parameter C for C-SVC, epsilon-SVR and nu-SVR. libsvm.cost=1.0 #The degree of the kernel. libsvm.degree=3 #The tolerance of the termination criterion. libsvm.eps=0.001 #The gamma to use, if 0 then 1/max_index is used. libsvm.gamma=0.0 #The type of kernel to use #linear: u' * v #poly: (gamma * u * v + coef0)^degree #rbf: exp(-gamma * |u-v|^2) #sigmoid: tanh(gamma * u' * v + coef0) libsvm.kernel.type=linear #The epsilon for the loss function in epsilon-SVR. libsvm.loss=0.1 #Whether to normalize the data. #true, false libsvm.normalize=false #The value of nu for nu-SVC, one-class SVM and nu-SVR. libsvm.nu=0.5 #Whether to generate probability estimates instead of -1/+1 for classification problems. #true, false libsvm.probability.estimates=true #Whether to use the shrinking heuristic. #true, false libsvm.shrinking=true #The weights to use for the classes, if empty 1 is used by default. #??? libsvm.weights= ############################################ ###### DECISION TREE PARAMETERS ####### #do pruning or not dtree.unpruned=false #confidence factor for pruning (smaller value forces more pruning) dtree.confidencefactor=0.3 #use REP (if false original C45 pruning is used) dtree.reducederrorpruning=false #minimum number of instances per leaf dtree.minnumobj=2 #number of data used for REP dtree.numfolds=5 ############################################ ###### RANDOM FOREST PARAMETERS ####### #number of randomly chosen features considered for splitting in each node randomforest.numfeatures=1 #number of random trees randomforest.numtrees=1 #minimum number of instances in a node to be split randomforest.leaf.minnum=15 #minimum precision of a leaf randomforest.leaf.precision=0.96 ############################################ ###### ANN PARAMETERS ####### #network parameters ann.hiddenlayers=10 ann.learningrate=0.3 ann.momentum=0.2 #normalize data ann.normalizeattributes=false #number of iterations for backpropagation ann.trainingtime=50 ####################################### ###### ADABOOST PARAMETERS ####### #classifiers: dtree, decisionstump adaboost.classifier=dtree #number of boosting iterations adaboost.numiterations=10 #if false, reweighting is used adaboost.useresampling=false #percentage of weight/number of instances used for each classifier adaboost.weightthreshold=100 ####################################### ###### MULTIBOOST PARAMETERS ####### #classifiers: dtree, decisionstump multiboost.classifier=dtree #number of boosting iterations multiboost.numiterations=10 #Sets the (approximate) number of subcommittees multiboost.numcommittees=10 #if false, reweighting is used multiboost.useresampling=false #percentage of weight/number of instances used for each classifier multiboost.weightthreshold=100 ####################################### ###### LOGITBOOST PARAMETERS ####### #Still to implement, with regression-type classifiers #classifiers: ????? logitboost.classifier=???? #number of boosting iterations logitboost.numiterations=10 #threshold on improvement in likelihood logitboost.likelihood.threshold=-1.7976931348623157E308 #Number of folds for internal cross-validation #(default 0 means no cross-validation is performed). logitboost.numfolds=0 #Number of runs for internal cross-validation. logitboost.numruns=1 #Shrinkage parameter (use small value like 0.1 to reduce overfitting) logitboost.shrinkage=1.0 #if false, reweighting is used logitboost.useresampling=false #percentage of weight/number of instances used for each classifier logitboost.weightthreshold=100