exp_pca: Exponential Family PCA via Alternating Minimization

In krisrs1128/expPCA: Exponential Family PCA

Description

General exponential-family PCA algorithm, within which different family-specific bregman losses can be input.

Usage

exp_pca(obj_fun, obj_grad_a, obj_grad_v, X, n_comp, n_cycle, n_iter, eps,
  lambda, mu0, verbose = FALSE)

Arguments

obj_fun	The main bregman loss function to attempt to minimize. See bern_breg() for an example of the required input / output format.
obj_grad_a	The derivative of the bregman loss with respect to the scores. See bern_brad_grad_a() for an example of the required input / output format.
obj_grad_v	The derivative of the bregman loss with respect to the loadings. See bern_brad_grad_a() for an example of the required input / output format.
X	The n x p binary matrix with samples along rows which we want to decompose using exponential family PCA.
n_comp	How many principal components should we return?
n_cycle	How many times should the iterative optimization pass through across each component?
n_iter	The maximum number of iterations to run the PCA.
eps	The convergence criterion. If the mean change in the scores is less than eps, we return.
lambda	The regularization parameter in the optimization.
mu0	The value to regularize towards.
verbose	Print iterations of procedure?

Value

A list containing the converged values of A and V.

References

Collins, Michael, Sanjoy Dasgupta, and Robert E. Schapire. "A generalization of principal components analysis to the exponential family." Advances in neural information processing systems. 2001.

krisrs1128/expPCA documentation built on May 20, 2019, 1:26 p.m.