Input
Training data	File should contain table data: lines for observations (samples) and columns for variables; columns should be separated by tabulation or user-defines sybol (;, etc); no missed data allowed. The training data is mandatory parameter.
Testing data	File should contain table data: lines for observations (samples) and columns for variables; columns should be separated by tabulation or user-defines sybol (;, etc); no missed data allowed. The training data is not mandatory parameter, if it is omitted, the testing will be performed on the training data.
Validating data	File should contain table data: lines for observations (samples) and columns for variables; columns should be separated by tabulation or user-defines sybol (;, etc); no missed data allowed. The validating data is not mandatory parameter, if it is omitted, the validating will be performed on the training data.
Structure	Recently obtained file with network parameters to start from this network. To continue training network from previously saved parameters the network structure file in MLP format can be specified. This parameter is optional. If it is not stated, the learning begins with random NN weights.
List of input variables	List of variables which serve as predictors for NN, the input of the neural network. Examples of input: 1;2;3-7;12; 1-12; ALL If 'Observation name' parameter is set on, variable list should not contain 1.
List of target variables	List of targets variables (to be predicted by neural network). Format of input: 1;2;3-7;12; ALL
Output
Status	Output file with the calculation status
Network structure	Output file with network structure and parameters in MLP format. This file can be used for prediction by neural network algorithm in snnbp.
Format in C-code file	Numerical format in C-code file. The format for weight data representation in C-code file. This is numerical (c-like, but without %) format for prediction output. Example: for .3 format the output will be presented as ...NNNN.NNN (where N - decimal numeral).
C-data	File to save neural network data as C function. The network parameters could be saved as C-code file. The parameter is optional. If it is not set, no C-code file will be generated.
Prediction output option	If this parameter is set ON, for each of the training/testing/validation file additional file with *.pred extention will be created containing predicted and observed values of the output variables.
Options
Significant digits	String in C-type format description (without %), examples: 5.3f; .5f; 3.0f
Check names of variables	Check names of variables from table first row: Take 1-st line in the table Take 1-st line in the table
Check names of samples	Check names of samples from table first column: Take 1-st line in the table Take 1-st line in the table
Column separation	Symbol for separation variables in line; by default tabulation and space.
Commentary line symbol	Commentary line symbol (if line starts from CommentSymbol, then this line is ignored)
Number of layers	Number of layers in the neural network, including input and layers
Hidden layers sizes	Number of neurons in each hidden layer separated by semicolon. Example: 10;3; for 10 neurons in 1st hidden layer and 3 neurons in the 2nd hidden layer.
Momentum	The momentum value
Learning rate	Learning rate
Gain	Gain, the slope of the sigmoid function in the non-linear transformation of the NN
Number of epochs	The number of epochs per trainig step in the learning process
Number of training steps	The number of training steps in the learning process
Threshold for large error deviation	This parameter specify the error threshold for learning stopping criteria. It meaning depend on the StopCriteria setting.
Stopping criteria	This parameter defines the criteria to stop learning process. Zero - if the error is 0 (default); NSteps - if the the error did not decreased last LargeErrDev steps; Barrier - if the error increases after reaching its minimum (min_err) and the error is min_err*LargeErrDev.
Error estimation source	This parameter specify on which data to estimate error for stopping criteria. Validating - for testing data; Training - for training data.
Sampling protocol	This parameter specify the sampling protocol. RandTime - random sampling and on-line training, random generator initialized from the timer; RandInit - the same as previous, but the initialization is from the internally defined integer; Sequentially - samples are presented sequentially from the data, batch trainin is performed.