[source]

Dense Random Projector#

The Dense Random Projector uses a random matrix sampled from a dense uniform distribution [-1, 1] to reduce the dimensionality of a dataset by projecting it onto a vector space of target dimensionality.

Interfaces: Transformer, Stateful

Data Type Compatibility: Continuous only

Parameters#

# Param Default Type Description
1 dimensions int The number of target dimensions to project onto.

Additional Methods#

Estimate the minimum dimensionality needed given total sample size and max distortion using the Johnson-Lindenstrauss lemma:

public static estimate(int $n, float $maxDistortion = 0.1) : int

Example#

use Rubix\ML\Transformers\DenseRandomProjector;

$dimensions = DenseRandomProjector::estimate(3500);

$transformer = new DenseRandomProjector($dimensions); // Using estimate

$transformer = new DenseRandomProjector(50); // Specifying dimensionality

References#

  • D. Achlioptas. (2003). Database-friendly random projections: Johnson-Lindenstrauss with binary coins.