Optional Syntax Elements for TD_SVM - Analytics Database

Database Analytic Functions

Deployment
VantageCloud
VantageCore
Edition
Enterprise
IntelliFlex
VMware
Product
Analytics Database
Release Number
17.20
Published
June 2022
Language
English (United States)
Last Update
2024-10-04
dita:mapPath
gjn1627595495337.ditamap
dita:ditavalPath
ayr1485454803741.ditaval
dita:id
jmh1512506877710
lifecycle
latest
Product Category
Teradata Vantageā„¢
OUT clause
Accepts the MetaInformationTable clause.
ModelType
Specifies the type of the analysis. Acceptable values are Regression, Classification.
Default: Classification
MaxIterNum
Specifies the maximum number of iterations (minibatches) over the training data batches. The value is a positive integer less than 10,000,000.
Default: 300
Epsilon
Specifies the epsilon threshold for Regression (the value of epsilon for epsilon_insensitive loss). Any difference between the current prediction and the correct label is ignored within this threshold. The value is numeric.

Epsilon is a numeric value that sets a threshold for the epsilon-insensitive loss, which means that any difference between the predicted value and the actual label within this threshold is ignored.

This means that the SVM algorithm allows for a margin of error, or "slack" when making predictions for regression problems. The value of epsilon determines the size of this margin of error and affects the tradeoff between the accuracy of the model and its ability to generalize to new data.

Default: 0.1
BatchSize
Specifies the number of observations (training samples) processed in a single minibatch per AMP. A value of 0 or higher than the number of rows on an AMP processes all rows on the AMP, such that the entire dataset is processed in a single iteration, and the algorithm becomes Gradient Descent. The value is a positive integer.
Default: 10
RegularizationLambda

Specifies the amount of regularization to be added. The higher the value, stronger the regularization. It must be a positive float value. A value of 0 means no regularization.

RegularizationLambda is a positive float value that specifies the amount of regularization to be added to the model. The higher the value of RegularizationLambda, the stronger the regularization, which means that the algorithm places a greater emphasis on preventing overfitting to the training data.

In addition to its use in regularization, RegularizationLambda is also used to compute the learning rate when learning rate is set to optimal. Learning rate is a parameter that controls the step size at each iteration of the optimization algorithm and setting it to optimal means that it is automatically determined based on the value of RegularizationLambda.

Default: 0.02
Alpha
Specifies the Elasticnet parameter for penalty computation. It is only effective when RegularizationLambda is greater than 0. The value represents the contribution ratio of L1 in the penalty. A value of 1.0 indicates L1 (LASSO) only, a value of 0 indicates L2 (Ridge) only, and a value between is a combination of L1 and L2. The value is a float value between 0 and 1.
Default: 0.15
IterNumNoChange
Specifies the number of iterations (minibatches) with no improvement in loss, including the tolerance to stop training. A value of 0 indicates no early stopping and the algorithm continues until MaxIterNum iterations are reached. The value is a positive integer.
Default: 50
Tolerance
Specifies the stopping criteria in terms of loss function improvement. Applicable when IterNumNoChange is greater than 0. The value is a positive integer.
Default: 0.001
Intercept
Specifies whether intercept is to be estimated based on whether data is already centered.
Default: True
ClassWeights
Specifies weights associated with classes. Only applicable for Classification. The format is 0:weight, 1:weight. For example, 0:1.0,1:0.5 gives twice the weight to each observation in class 0 compared to class 1. If the weight of a class is omitted, it is assumed to be 1.0.
Default: 0:1.0, 1:1.0
LearningRate
Specifies the learning rate algorithm. Learning rates are:
  • Constant
  • InvTime
  • Optimal
  • Adaptive
Default: InvTime for Regression, Optimal for Classification
InitialEta
Specifies the initial value of eta for learning rate. For LearningRate set to constant, this value is the learning rate for all iterations.

The learning rate controls how much the SVM algorithm adjusts the weights of the model during training. If the learning rate is too low, the model may converge slowly. If the learning rate is too high, the model may fail to converge at all.

The choice of learning rate can have a significant impact on the accuracy and speed of the SVM algorithm, so it is an important parameter to consider when training a model.

Default: 0.05
DecayRate
Specifies the decay rate for learning rate. Only applicable for learning rates invtime and adaptive.
Default: 0.25
DecaySteps
Specifies the number of iterations without decay for the adaptive learning rate. The learning rate changes by decay rate after this many iterations.
Default: 5
Momentum
Specifies the value to use for momentum learning rate optimizer. A larger value indicates higher momentum contribution. A value of 0 means momentum optimizer is disabled. A value between 0.6-0.95 is recommended. The value is a positive float between 0 and 1.
Default: 0
Nesterov
Specifies whether to apply the Nesterov optimization to Momentum optimizer or not. Only applicable when Momentum is greater than 0.
Default: False
LocalSGDIterations
Specifies the number of local iterations to be used for Local SGD algorithm. A value of 0 means Local SGD is disabled. A value higher than 0 enables Local SGD and multiple, equal to the value supplied by the user. With Local SGD algorithm, the recommended values for arguments are:
  • LocalSGDIterations: 10
  • MaxIterNum: 100
  • BatchSize: 50
  • IterNumNoChange: 5
The value is a positive integer.
Default: 0