RandPro (version 0.1.0)

form_sparse_matrix: Form Sparse Random Matrix

Description

The purpose of sparse matrix is to reduce the computational complexity even more when comparing to gaussian matrix. The sparse matrix can be generated using a simpler approach by filling with zero's or one's to reduce the computation. This function supports 3 methods that are used to generate the sparse matrix.

Usage

form_sparse_matrix(n_rows, n_cols, JLT, eps = 0.1, method = "achlioptas")

Arguments

n_rows

- number of rows in the sample

n_cols

- number of columns in the sample

JLT

- Boolean to set JL transform (TRUE or FALSE)

eps

- error tolerance level

method

- method to generate projection matrix

Value

Random Sparse Matrix

Details

The 3 types of methods used in the function are

1. "probability" - In this method, the sparse matrix was generated using the equal probability distribution with the elements [-1, 1].

2. "achlioptas" - Achlioptas matrix is easy to generate and also the 2/3rd of the matrix was filled with zero which makes it as more sparse and cut-off the 2/3rd computation.

3. "li" - This method generalizes the achlioptas method and generate very sparse random matrix to improve the computational speed up of random projection.

References

[1] Ping Li, Trevor J. Hastie, and Kenneth W. Church, "Very sparse random projections(2006)".

[2] D. Achlioptas, "Database-friendly random projections(2002)"

Examples

# NOT RUN {
#load library
library(RandPro)

# Matrix generated with deafult method "achlioptas" without JLT
mat <- form_sparse_matrix(n_rows = 300,n_cols = 1000,JLT = FALSE)

#create matrix with JLT and default method
mat <- form_sparse_matrix(n_rows = 250,n_cols = 1000000,JLT = TRUE,eps = 0.5)

#create matrix with JLT and probability method
mat <- form_sparse_matrix(250,1000000,JLT = TRUE,eps = 0.5,method = "probability")

#create matrix with JLT and li method
mat <- form_sparse_matrix(n_rows = 250,n_cols = 1000000,JLT = TRUE,eps = 0.5,method = "li")

# }