Introduction

The filematrix package provides functions to create and access large matrices stored in files, not computer memory. Filematrices can be as large as the storage allows. The package has been tested on matrices multiple terabytes in size.

Creating a filematrix

File matrices can be created using functions fm.create, fm.create.from.matrix, or fm.create.from.text.file:

library(filematrix)
fm = fm.create(filenamebase = "fmat", nrow = 200, ncol = 200, type = "double")

The code above creates two files: fmatrix.bmat which stores the filematrix values, and fmatrix.desc.txt which stores the filematrix description, such as dimensions, data type, and data type size.

Here is the content of the description file:

cat(readLines("fmat.desc.txt"), sep = "\n")

# Information file for R filematrix object
nrow=200
ncol=200
type=double
size=8

Reading from and writing to a filematrix

The elements of a filematrix can be read and written to using the same syntax as is used for regular R matrices. Any changes to a filematrix are written to the .bmat file without extra buffering.

fm[1:3, 1:2] = 1:6
fm[1:4, 1:3]

##      [,1] [,2] [,3]
## [1,]    1    4    0
## [2,]    2    5    0
## [3,]    3    6    0
## [4,]    0    0    0

colSums(fm[,1:4])

## [1]  6 15  0  0

Elements of a filematrix can also be accessed as elements of a vector (in which elements proceed sequentially down columns stacked 1:n). Thus, as fm has nrow(fm) rows, fm[1,2] accesses the same element as fm[nrow(fm)+1].

fm[1:4]

## [1] 1 2 3 0

fm[nrow(fm)+1:4]

## [1] 4 5 6 0

Row and column names

File matrices can also have row and column names, like regular R matrices.

colnames(fm) = paste0("Col", 1:ncol(fm))
rownames(fm) = paste0("Row", 1:nrow(fm))

The row and column names of the filematrix fm are stored in fmatrix.nmsrow.txt and fmatrix.nmscol.txt respectively.

Closing filematrices

An open filematrix object can be closed with close function. This closes the internal file handle (connection). Closing filematrix objects is optional, changes would not be lost if the object is not closed.

close(fm)

Open or load an existing filematrix

An existing filematrix can be opened with fm.open.

fm = fm.open(filenamebase = "fmat", readonly = FALSE)

To prevent any changes to the values of the filematrix set readonly = TRUE.

An existing filematrix that would fit in memory can be loaded fully with fm.load

mat = fm.load("fmat")

By column access is faster then by rows

The values of a filematrix are stored by columns, as with regular R matrices:

matrix(1:12, nrow = 3, ncol = 4)

##      [,1] [,2] [,3] [,4]
## [1,]    1    4    7   10
## [2,]    2    5    8   11
## [3,]    3    6    9   12

Thus, access to a filematrix values by columns is much faster than access by rows:

timerow = system.time( { sum(fm[1:10, ]) } )[3]
timecol = system.time( { sum(fm[ ,1:10]) } )[3]
cat("Reading ", nrow(fm)*10, " values from 10 columns takes ", timecol, " seconds", "\n",
    "Reading ", ncol(fm)*10, " values from 10 rows takes ",    timerow, " seconds", "\n", sep = "")

## Reading 2000 values from 10 columns takes 0.001 seconds
## Reading 2000 values from 10 rows takes 0.01 seconds

The performance difference may not be observed on small matrices, as in this example. Change the size from 200 x 200 to 10,000 x 10,000 to see the difference (it is at least hundred fold).

Appending columns

Unlike with regular R matrices, columns can be appended to the right side of a filematrix with very little computational overhead.

dim(fm)

## [1] 200 200

fm$appendColumns(nrow(fm):1)
fm$appendColumns(1:nrow(fm))
dim(fm)

## [1] 200 202

fm[nrow(fm)+(-1:0), ncol(fm)+(-1:0)]

##      [,1] [,2]
## [1,]    2  199
## [2,]    1  200

Deleting filematrix files

If you no longer need a filematrix and want to delete its files from the hard drive and close the object, please use closeAndDeleteFiles()

closeAndDeleteFiles(fm)

Version information

sessionInfo()

## R version 4.4.1 (2024-06-14)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 24.04.1 LTS
## 
## Matrix products: default
## BLAS:   /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3 
## LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so;  LAPACK version 3.12.0
## 
## locale:
##  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=C              
##  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
##  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
## 
## time zone: Etc/UTC
## tzcode source: system (glibc)
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
## [1] filematrix_1.3 knitr_1.48    
## 
## loaded via a namespace (and not attached):
##  [1] digest_0.6.37     R6_2.5.1          fastmap_1.2.0     xfun_0.49        
##  [5] maketools_1.3.1   cachem_1.1.0      htmltools_0.5.8.1 rmarkdown_2.28   
##  [9] buildtools_1.0.0  lifecycle_1.0.4   cli_3.6.3         sass_0.4.9       
## [13] jquerylib_0.1.4   compiler_4.4.1    sys_3.4.3         tools_4.4.1      
## [17] evaluate_1.0.1    bslib_0.8.0       yaml_2.3.10       jsonlite_1.8.9   
## [21] rlang_1.1.4