JACVANN.bs Reference

command range of values default value description
OutputFile filename - filename to save the output of JACVANN
DataSetFile path to .cds file - path to the .cds file containing trainings data set
Topo x,y,z[,y...] - topology of the network. integers seperated by decimal point
OutputLayerOutputFunction 'outputLinear', 'outputSigmoid', 'outputHybridMag' or 'outputHybridSquareDiff' 'outputLinear' optional: output function of output layer
InputEncoding 'noEncoding', 'polarEncoding' or 'complexNullEncoding' 'noEncoding' optional: encoding of input vector
InitRange [x,y] [-0.3,0.3] optional: range of floating point numbers seperated by decimal point
InitType 'initRN' or 'initNW' 'initRN' optional: initialization method: initRN (random initialization), initNW (initialization by Nguyen and Widrow)
TrainAlgo 'BackProp' or 'RProp' - training function
LearningRate 0.0 - 1.0 0.1 optional: learning rate
EtaN 0.0 - 2.0 0.5 optional: reducing factor RProp
EtaP 0.0 - 2.0 1.2 optional: amplification factor RProp
DeltaMin 0.0 - 1.0 1.0E-6 optional: minimal modification RProp
DeltaMax 1.0 - 1.7E+308 50 optional: maximalb modification RProp
TrainType 'TrainTestRun' or 'CrossValidation' 'TrainTestRun' type of run: simple training run of cross validation
TestDataAmount 0.0 - 1.0 0.25 optional: use only if TrainType = 'TrainTestRun'
CrossValidationParts 1 - 'amount of data' 4 optional: use only if TrainType = 'CrossValidation'
ErrorFunction 'RMSD', 'MSE', 'ThreshSR', 'ThreshSSU' or 'ThreshSD' 'RMSD' error function to use
Threshold 0.0 - 1.7E+308 0.5 if errorfunction equals 'Thres' the threshold is set here
Epochs > 0 50000 optional: number of epochs to train
BadRuns >= 0 1000 optional: maximum number of bad runs
ErrorGoal >= 0.0 1.0E-4 optional: error to reach
SaveNet 'true' or 'false' 'false' save best network of a run to a file
NetFile path/filename 'netFile.cvn' necessary, if SaveNet = 'true'. Should change after every run so that old data will not get lost
Print at will - optional: writes the text to console window and, if set, to the output file
Train - - trains the network with given parameters
End - - ends the programm