Lodestar.Decomposition 0.1.0. This page is frozen at that release. Read the current documentation for what main says now. A link to a decision or a migration page follows main, and leaves the archive.

TruncatedSvd.Transform

Projects rows onto the fitted components — X · Componentsᵀ, and nothing else.

public double[] Transform(CsrMatrix matrix)

Parameters — matrix holds the rows to project, and must have exactly FeatureCount columns. It does not have to be the matrix that was fitted, and usually is not.

Returns — double[], row-major and ComponentCount wide: matrix.RowCount × ComponentCount values, row i's coordinates starting at i * ComponentCount.

Exceptions — ArgumentNullException when matrix is null. ArgumentException when matrix does not have FeatureCount columns.

Example — the same four documents, projected onto the two components fitted from them.

using Lodestar.Abstractions;
using Lodestar.Decomposition;

CsrMatrix corpus = new(
    4, 3,
    [3.0, 1.0, 2.0, 1.0, 1.0, 4.0, 2.0, 3.0],
    [0, 1, 0, 2, 1, 2, 0, 2],
    [0, 2, 4, 6, 8]);

TruncatedSvd model = TruncatedSvd.Fit(corpus, 2);
double[] projected = model.Transform(corpus);

int values = projected.Length;                             // => 8
double firstDocument = Math.Round(projected[0], 3);        // => 1.672

Remarks — this is not U · Σ. The two agree for an exact SVD and differ by the randomized solver's approximation error for this one, which on a real corpus is visible in the third decimal rather than the last bit. scikit-learn's TruncatedSVD.transform computes the projection, so this does too, and ExplainedVariance is measured on it for the same reason.

Nothing is centred and nothing is scaled, here or in the fit, so a row of zeros projects to the origin and a document twice as long projects twice as far. Normalize the rows before fitting if that is not what you want — CsrMatrix.NormalizeRows does it in place.

Applies to — net10.0, netstandard2.0.

See also — TruncatedSvd.Fit, the Python equivalence table.