Replaced dgCMatrix with CsparseMatrix in tests.

DESCRIPTION

@@ -1,7 +1,7 @@
 Encoding: UTF-8
 Type: Package
 Package: fastTopics
-Version: 0.6-138
+Version: 0.6-139
 Date: 2022-08-24
 Title: Fast Algorithms for Fitting Topic Models and Non-Negative
     Matrix Factorizations to Count Data

R/fit_poisson_nmf.R

@@ -428,7 +428,7 @@ fit_poisson_nmf <- function (X, k, fit0, numiter = 100,
       method.text <- "CCD"
     cat(sprintf("Running %d %s updates, %s extrapolation ",numiter,
         method.text,ifelse(control$extrapolate,"with","without")))
-    cat("(fastTopics 0.6-138).\n")
+    cat("(fastTopics 0.6-139).\n")
   }
 
   # INITIALIZE ESTIMATES

tests/testthat/test_de_analysis.R

@@ -86,7 +86,7 @@ test_that(paste("All variants of fit_poisson_models should produce the",
   out <- add_pseudocounts(X,s*L,0.1)
   X   <- out$X
   L   <- out$L
-  Y   <- as(X,"dgCMatrix")
+  Y   <- as(X,"CsparseMatrix")
 
   # Fit the univariate Poisson models using glm and scd.
   F1 <- fit_poisson_models(X,L,method = "glm")
@@ -115,7 +115,7 @@ test_that(paste("de_analysis with and without multithreading, using a",
   k   <- 4
   dat <- simulate_multinom_gene_data(n,m,k,sparse = FALSE)
   X   <- dat$X
-  Y   <- as(X,"dgCMatrix")
+  Y   <- as(X,"CsparseMatrix")
   L   <- dat$L
 
   # Run de_analysis twice, using the single-threaded computations (nc
@@ -161,7 +161,7 @@ test_that(paste("de_analysis with and without multithreading, using a",
   k   <- 4
   dat <- simulate_multinom_gene_data(n,m,k,sparse = FALSE)
   X   <- dat$X
-  Y   <- as(X,"dgCMatrix")
+  Y   <- as(X,"CsparseMatrix")
   L   <- dat$L
 
   # Run de_analysis twice, using the single-threaded computations (nc

tests/testthat/test_fit_multinom_model.R

@@ -9,7 +9,7 @@ test_that("fit_multinom_model gives correct factor estimates",{
   k   <- 3
   out <- simulate_toy_gene_data(n,m,k,s = 1000)
   X   <- out$X
-  Y   <- as(X,"dgCMatrix")
+  Y   <- as(X,"CsparseMatrix")
 
   # Force "hard" topic assignments.
   cluster <- factor(apply(force_hard_topic_assignments(out$L),1,which.max))

tests/testthat/test_fit_poisson_nmf.R

@@ -87,7 +87,7 @@ test_that(paste("multiplicative and EM updates produce same result, and",
                             control = list(numiter = 1,nc = nc)))
 
   # Store the counts as a sparse matrix.
-  Y <- as(X,"dgCMatrix")
+  Y <- as(X,"CsparseMatrix")
 
   # Run 20 EM updates a third time, this time using the sparse counts
   # matrix.
@@ -165,7 +165,7 @@ test_that(paste("ccd and scd updates produce the same result, and",
                             control = list(numiter = 1,nc = 1)))
 
   # Redo the SCD updates with a sparse matrix.
-  Y <- as(X,"dgCMatrix")
+  Y <- as(X,"CsparseMatrix")
   capture.output(
       fit3 <- fit_poisson_nmf(Y,fit0 = fit0,numiter = numiter,method = "scd",
                               control = list(numiter = 1,nc = 1)))
@@ -492,7 +492,7 @@ test_that("Fixed factors and loadings to not change (aside from rescaling)",{
   k   <- 3
   out <- generate_test_data(n,m,k)
   X   <- out$X
-  Y   <- as(X,"dgCMatrix")
+  Y   <- as(X,"CsparseMatrix")
 
   # Check that the factors remain the same (up to a rescaling) when
   # update.factors is NULL.

tests/testthat/test_likelihood.R

@@ -12,7 +12,7 @@ test_that(paste("R and Rcpp versions of cost function return same result",
     X   <- out$X
     F   <- out$F
     L   <- out$L
-    Y   <- as(X,"dgCMatrix")
+    Y   <- as(X,"CsparseMatrix")
 
     # Compute the loss function.
     f1 <- cost(X,L,t(F),version = "R")
@@ -43,7 +43,7 @@ test_that(paste("loglik_poisson_nmf gives correct result for sparse and",
     # Compute the log-likelikhood.
     f1 <- loglik_poisson_nmf_with_dpois(X,fit)
     f2 <- loglik_poisson_nmf(X,fit,e = 0)
-    f3 <- loglik_poisson_nmf(as(X,"dgCMatrix"),fit,e = 0)
+    f3 <- loglik_poisson_nmf(as(X,"CsparseMatrix"),fit,e = 0)
     names(f1) <- rownames(X)
 
     # The likelihood calculations should all be the same.
@@ -68,7 +68,7 @@ test_that(paste("loglik_multinom_topic_model gives correct result for",
     # Compute the log-likelikhood.
     f1 <- loglik_multinom_topic_model_with_dmultinom(X,poisson2multinom(fit))
     f2 <- loglik_multinom_topic_model(X,fit,e = 0)
-    f3 <- loglik_multinom_topic_model(as(X,"dgCMatrix"),fit,e = 0)
+    f3 <- loglik_multinom_topic_model(as(X,"CsparseMatrix"),fit,e = 0)
 
     # Compute the multinomial log-likelihood a different way: first
     # compute the Poisson log-likelihood, then subtract out the
@@ -98,7 +98,7 @@ test_that(paste("deviance_poisson_topic_nmf gives correct result for sparse",
     # Compute the deviances.
     d1 <- deviance_poisson_nmf_with_poisson(X,fit)
     d2 <- deviance_poisson_nmf(X,fit,e = 0)
-    d3 <- deviance_poisson_nmf(as(X,"dgCMatrix"),fit,e = 0)
+    d3 <- deviance_poisson_nmf(as(X,"CsparseMatrix"),fit,e = 0)
 
     # The deviance calculations should all be the same.
     expect_equal(d1,d2)
@@ -117,7 +117,7 @@ test_that("poisson_nmf_kkt gives same result for sparse and dense matrices",{
 
   # Compute the KKT residuals, and check that they are the same.
   out1 <- poisson_nmf_kkt(X,F,L)
-  out2 <- poisson_nmf_kkt(as(X,"dgCMatrix"),F,L)
+  out2 <- poisson_nmf_kkt(as(X,"CsparseMatrix"),F,L)
   expect_equal(out1,out2,tolerance = 1e-15,scale = 1)
 })
 

tests/testthat/test_poisson2multinom.R

@@ -34,7 +34,7 @@ test_that("multinom2poisson recovers original Poisson NMF model fit",{
   fit2a$s <- NULL
   fit3 <- multinom2poisson(fit2)
   fit4 <- multinom2poisson(fit2a,X)
-  fit5 <- multinom2poisson(fit2a,as(X,"dgCMatrix"))
+  fit5 <- multinom2poisson(fit2a,as(X,"CsparseMatrix"))
   Y1   <- with(fit1,tcrossprod(L,F))
   Y3   <- with(fit3,tcrossprod(L,F))
   Y4   <- with(fit4,tcrossprod(L,F))