7.00.02 - NeuralNet Arguments - Aster Analytics

Teradata Aster® Analytics Foundation User GuideUpdate 2

Product
Aster Analytics
Release Number
7.00.02
Release Date
September 2017
Content Type
Programming Reference
User Guide
Publication ID
B700-1022-700K
Language
English (United States)
InputTable
Specifies the table containing the input data to be trained.
OutputTable
Specifies the table to which to output the trained network weight data.
WeightTable
[Optional] Specifies the table that lists the starting values for the neural network weights. Default behavior: The function assigns the initial weights for the neural network randomly.
InputColumns
Specifies the name of the columns of the InputTable that contains the numerical predictor variables (x1, x2, x3, and so on).
ResponseColumns
Specifies the names of the input table columns of that contain the numerical dependent variables (y1, y2, y3, and so on).
GroupByColumns
[Optional] Specifies the weight table columns in which to output different neural networks for different groups.
HiddenLayers
[Optional] Specifies the number of hidden neurons in each layer, from left to right. The hidden_layers must be a nonnegative INTEGER. Default: 1 layer, 1 neuron. For example, HiddenLayers('5','5') produces a 3-layer network with 5 neurons in each hidden layer, while HiddenLayers('3') produces the network shown in Introduction to Neural Networks.
Threshold
[Optional] Specifies the threshold for the partial derivatives of the error function as stopping criteria. The threshold must be a DOUBLE value. Default: 0.01.
MaxIterNum
[Optional] Specifies the maximum number of iterations for the training of the neural network. The max_iterations must be an INTEGER value. Default: 1.
LearningRate
[Optional] Specifies the learning rate used by traditional backpropagation. Default: 0.001.
ActivationFunction
[Optional] Specifies the name of the differentiable function that the function applies to the result of the cross-product of the neurons and the weights.:
  • 'logistic': logistic (Default)
  • 'tanh': hyperbolic tangent
ErrorFunction
[Optional] Specifies the name of the differentiable function with which the function calculates the error:
  • 'sse': sum of squared errors (Default)
  • 'ce': cross-entropy
Algorithms
[Optional] Specifies the algorithm type that the function uses to calculate the neural network, 'backprop'.
LinearOutput
[Optional] Specifies whether to apply the ActivationFunction to the output neurons. Default: 'true'.
OverwriteOutput
[Optional] Specifies whether to overwrite the output table. Default: 'false'.
Seed
[Optional] Specifies the seed with which to initialize the model, an INTEGER. Given the same seed, cluster configuration, and input table, the function generates the same model. Default behavior: The function initializes the model randomly.