TD_RandomProjectionFit Syntax Elements - Teradata VantageCloud Lake

TargetColumns

[Required]: Specify the input table columns for dimensionality reduction.

NumComponents

[Required]: Specify the target dimension (number of features) on which the data points from the original dimension are projected.

The NumComponents value cannot be greater than the original dimension (number of features) and must satisfy the Johnson-Lindenstrauss Lemma result. The minimum value allowed for the NumComponents argument is calculated using the TD_RandomProjectionMinComponents function.

Seed

[Optional]: Specify the random seed the algorithm uses for repeatable results. The algorithm uses the seed to generate a random projection matrix. The seed must be a non-negative integer value.

Default Value: The Random Seed value is used for generating a random projection matrix, and hence the output is non-deterministic.

Epsilon

[Optional]: Specify a value to control distortion introduced while projecting the data to a lower dimension. The amount of distortion increases if you increase the value.

Default Value: 0.1

Allowed Values: Between 0 and 1

ProjectionMethod

[Optional]: Specify the method name for generating the random projection matrix.

Default Value: GAUSSIAN

Allowed Values: [GAUSSIAN, SPARSE]

Density

[Optional]: Specify the approximate ratio of non-zero elements in the random projection matrix when SPARSE is used as the projection method.

Default Value: 0.33333333

Allowed Values: 0 < Density <= 1

OutputFeatureNamesPrefix

[Optional]: Specify the prefix for the output column names.

Default Value: td_rpj_feature