DimensionReducerFunction

Status: Stable

documented, exercised by the test suite and/or worked examples, with no known limitations recorded.

Description

DimensionReducerFunction[method, {means, loadings...}, featureCount, reducedDimension]

is the reusable reducer DimensionReduction returns. Apply it to a feature vector, or to a matrix of them, to project into the reduced space.

Examples (2)

Every input below was run against the current Mathilda build and its output recorded.

Applications (2)

Returns a reusable reducer

In[1]:= Head[DimensionReduction[{{0, 0}, {1, 1}, {2, 2}, {3, 3}}, 1]]
Out[1]= DimensionReducerFunction

Query a property

In[2]:= r = DimensionReduction[{{0, 0}, {1, 1}, {2, 2}, {3, 3}}, 1]; r["ReducedDimension"]
Out[2]= 1

Implementation notes

Definition. DimensionReducerFunction[method, {means, loadings...}, featureCount, reducedDimension] is the reusable projector that DimensionReduction[data, k] returns. It is an inert head (no builtin, just a docstring and the Protected attribute), produced by DimensionReduction and consumed by predict.c when applied.

Representation. A four-argument EXPR_FUNCTION: the method, a parameter block holding the training-column means and the projection loadings (the retained principal axes), the input featureCount, and reducedDimension — the target dimension k. New rows are centred on the stored training means before projection, which is what makes projections comparable across batches.

Usage & limits. Protected. Apply it to one feature vector, or to a matrix of vectors, to project into the reduced space; it also answers "Method", "FeatureCount" and "ReducedDimension". The difference from DimensionReduce[data, k] — which returns the reduced training data — is that this object generalises to data it was not trained on. It holds only the means and loadings, so it cannot reconstruct the original rows.

Attributes: Protected.

References

• Source: src/ml/predict.c
• Specification index: Mathilda_spec.md
• Tests: tests/test_ml_predict.c

Notes & additional examples

Notes

DimensionReducerFunction[method, {means, loadings...}, featureCount, reducedDimension] is the projector that DimensionReduction[data, k] returns — a reusable object, not a user-written head. Apply it to a feature vector, or to a matrix of vectors, to project into the reduced k-dimensional space; it also answers "Method", "FeatureCount" and "ReducedDimension".

New rows are centred on the stored training-column means before projection, which is what makes projections comparable across batches. That is the difference from DimensionReduce[data, k], which returns the reduced training data: the DimensionReducerFunction generalises to data it was not trained on. It is Protected.