dgpage_discriminant: Fit DGPAGE and return a discriminant_projector

In bbuchsbaum/discursive: What the Package Does (One Line, Title Case)

Fit DGPAGE and return a discriminant_projector

Description

This function wraps dgpage_fit but returns a discriminant_projector (subclass "dgpage_projector") so that you can use the same predict interface as for other discriminant methods.

Usage

dgpage_discriminant(
  X,
  y,
  S = NULL,
  D = NULL,
  r = 2,
  alpha = 0.001,
  beta = 1e-05,
  maxiter = 20,
  tol = 1e-05,
  verbose = TRUE,
  q = 1.5
)

Arguments

X	A numeric matrix of size n x d (training samples).
y	A factor (or coercible to factor) of length n (class labels).
S	The n x n similarity graph. If NULL, we compute it from X,y.
D	The n x n diversity graph. If NULL, we compute it from X,y.
r	The target dimension (number of discriminant directions).
alpha, beta	Hyperparameters in Eq. (12).
maxiter, tol, verbose	Passed to dgpage_fit.
q	If S or D is NULL, we use q in compute_similarity_graph. Default 1.5.

Details

Algorithm:

1. If S or D is NULL, we build them from X, y via compute_similarity_graph and compute_diversity_graph.

2. We call dgpage_fit to jointly learn the adjacency graph W and projection matrix P.

3. We create a discriminant_projector object with:

   • v = result$P (the loadings),

   • s = X %*% result\$P (the training scores),

   • sdev a vector of stdev per dimension,

   • labels = y.

   We store W and other metadata in the returned object.

NOTE: The dgpage_predict for 1-NN classification can now be replaced by a new S3 predict method that lives under dgpage_projector.

Value

An object of class c("dgpage_projector","discriminant_projector","bi_projector",...) containing:

• v : the d x r loadings

• s : the n x r scores

• sdev : numeric vector of length r with stdevs of columns of s

• labels : the training labels

• W : the learned n x n adjacency graph

• objective: final objective value

• alpha,beta : stored hyperparameters

Examples

## Not run: 
  X <- matrix(rnorm(100*5), 100, 5)
  y <- factor(rep(1:2, each=50))
  dp <- dgpage_discriminant(X, y, r=2)
  preds_class  <- predict(dp, new_data=X, method="lda", type="class")
  preds_1nn    <- predict(dp, new_data=X, method="1nn")
  mean(preds_class == y)
  mean(preds_1nn   == y)

## End(Not run)

bbuchsbaum/discursive documentation built on April 14, 2025, 4:57 p.m.