README.md

In emauryg/STRAND_R: Structural TensoR ANalysis and Decomposition (STRAND)

STRAND_R

Structural TensoR Analsyis and Decomposition

Installation

For the most recent stable version library(devtools); install_github("emauryg/STRAND_R");

For the development version `library(devtools); install_github("emauryg/STRAND_R",ref="development")

Quickstart

library(MCMCpack); library(mvtnorm); library(NMF); library(clue); library(nnet); 

library(batch); library(Matrix); library(torch); library(tidyverse)

library(strandR)

if (cuda_is_available()) {
   device <<- torch_device("cuda:0")
} else {
   device <<- torch_device("cpu")
}

D=100 ## number of samples
K=5 ## number of latent factors (signatures)
V=96 ## number of trinucleotide contexts
p=5 ## number of sample covariates

set.seed(777)
torch_manual_seed(777)

truth_vals <- generate_data(V,K,D,p)

count_matrix = truth_vals$count_matrix
xmat = truth_vals$X

## Initialize 
init_pars <- NMFinit(count_matrix,xmat,K, max_iter=10000) 

## Fit model
mod0 = runEM(init_pars, count_matrix, X=xmat)

## Plot 96 trinucleotide context
make_96plot(mod0,model_type = "STRAND")

## Plot factors
factor_plots = plot_factors(mod0)

GPU acceleration

Using torch STRAND can make use of GPUs if the current device has that infrastructure available.

if (cuda_is_available()) {
   device <<- torch_device("cuda:0")
} else {
   device <<- torch_device("cpu")
}

This can provide significant speed gains compared to just using cpu.

TODO:

• Need to incorporate package names of none base R for ease of use.

emauryg/STRAND_R documentation built on Dec. 20, 2021, 4:22 a.m.