Gaussian Random Projector#
Random Projection is a dimensionality reduction technique based on the Johnson-Lindenstrauss lemma. It uses random matrices to project feature vectors onto a target number of dimensions. The Gaussian Random Projector utilizes a random matrix sampled from a smooth Gaussian distribution which projects samples onto a spherically random hyperplane through the origin.
Interfaces: Transformer, Stateful, Persistable
Data Type Compatibility: Continuous only
Parameters#
# | Name | Default | Type | Description |
---|---|---|---|---|
1 | dimensions | int | The number of target dimensions to project onto. |
Example#
use Rubix\ML\Transformers\GaussianRandomProjector;
$transformer = new GaussianRandomProjector(100);
Additional Methods#
Estimate the minimum dimensionality needed to satisfy a max distortion constraint with n samples using the Johnson-Lindenstrauss lemma:
public static minDimensions(int $n, float $maxDistortion = 0.5) : int
use Rubix\ML\Transformers\GaussianRandomProjector;
$dimensions = GaussianRandomProjector::minDimensions(5000, 0.2);
Last update:
2021-01-23