visentin-insight / l1-pcahp Goto Github PK

This repository contains all the code used for the "Computational Results" of the paper 

"Robust Principal Component Analysis by Reverse Iterative Linear Programming" 
Andrea Visentin, Steven Prestwich and Armagan Tarim

## FILE LIST ##

./data:          			## this directory contains all the dataset. You must 
					## retrive them 
					## from http://archive.ics.uci.edu/ml/index.html


./pca:					## this directory contains the implementation of PCA

	l1pcahp.R			## the solution we introduce in the paper
					## more detailed description at the end of the page
					## implementation op PCA in R
	l2pca.R				## a container function


./src:					## this directory contains shared functions
	
	classifier.R			## 10cv 1nn classification used in Section 4.3
	dataset_manager.R		## load, clean and format the data for Sections 4.2, 
					## 4.3 and 4.5
	syntetic_data_generator.R	## data generator for test in Section 4.1


./test:

	classification.R		## test present in Section 4.3
	initialization_comparation.R	## test present in Section 4.5
	outliers_reconstruction.R	## test present in Section 4.2, reconstruction error
	outliers_variance.R		## test present in Section 4.5, variance of the projected
					## points
	syntetic.R			## test present in Section 4.1
	time_comparison.R		## test present in Section 4.4

## R PACKAGE REQUIRED ##	
	-pcaL1
	-pcaPP
	-MASS
	-matrixStats
	-wle
	-class
	-pdist


## ADDITIONAL SOFTWARE REQUIRED ##

MIP SOLVERS:
	- COIN-OR Clp (>= 1.12.0)
	- IBM ILOG CPLEX (>= 12.1)






## Description for function l1pcahp

This method implement the L1-PCAhp solution introduced in :
"Robust Principal Component Analysis by Reverse Iterative Linear Programming"
Andrea Visentin, Steven Prestwich and Armagan Tarim


## Usage

l1pcahp(X, projDim=1, center=TRUE, projPoints=FALSE, scores=TRUE, initialize="l2pca", threshold=1.0)


## Arguments

X			data, must be in matrix or table formdata, must be in matrix or table form
projDim 		number of dimensions to project data into, must be an integer, default is 1
center			whether to center the data using the median, default is TRUE
projPoints		whether to calculate the projected points in terms of original coordinates 
			(reconstruction),default is FALSE
scores			whether to return the matrix of scores, default is TRUE
initialize		method for initial guess for loadings matrix. Options are: "l2pca" - use 
			traditional PCA/SVD, "random" - use a randomly-generated matrix 
threshold		sets the convergence tolerance for the algorithm, default is 0.001