## For both 'Extract' ("[") and 'Replace' ("[<-") Method testing library(Matrix) source(system.file("test-tools.R", package = "Matrix"))# identical3() etc options(verbose = TRUE)# to show message()s ### Dense Matrices m <- Matrix(1:28 +0, nrow = 7) validObject(m) stopifnot(identical(m, m[]), identical(m[2, 3], 16), # simple number identical(m[2, 3:4], c(16,23))) # simple numeric of length 2 m[2, 3:4, drop=FALSE] # sub matrix of class 'dgeMatrix' m[-(4:7), 3:4] # ditto; the upper right corner of 'm' ## rows or columns only: m[1,] # first row, as simple numeric vector m[,2] # 2nd column m[,1:2] # sub matrix of first two columns m[-(1:6),, drop=FALSE] # not the first 6 rows, i.e. only the 7th m[integer(0),] #-> 0 x 4 Matrix m[2:4, numeric(0)] #-> 3 x 0 Matrix ## logical indexing stopifnot(identical(m[2,3], m[(1:nrow(m)) == 2, (1:ncol(m)) == 3]), identical(m[2,], m[(1:nrow(m)) == 2, ]), identical(m[,3:4], m[, (1:4) >= 3])) ## dimnames indexing: mn <- m dimnames(mn) <- list(paste("r",letters[1:nrow(mn)],sep=""), LETTERS[1:ncol(mn)]) mn["rd", "D"] stopifnot(identical(mn["rc", "D"], mn[3,4]), mn[3,4] == 24, identical(mn[, "A"], mn[,1]), mn[,1] == 1:7, identical(mn[c("re", "rb"), "B"], mn[c(5,2), 2]) ) mo <- m m[2,3] <- 100 m[1:2, 4] <- 200 m[, 1] <- -1 m[1:3,] m. <- as.matrix(m) ## m[ cbind(i,j) ] indexing: ij <- cbind(1:6, 2:3) stopifnot(identical(m[ij], m.[ij])) ## testing operations on logical Matrices rather more than indexing: g10 <- m [ m > 10 ] stopifnot(18 == length(g10)) stopifnot(10 == length(m[ m <= 10 ])) sel <- (20 < m) & (m < 150) sel.<- (20 < m.)& (m.< 150) nsel <-(20 >= m) | (m >= 150) (ssel <- as(sel, "sparseMatrix")) stopifnot(is(sel, "lMatrix"), is(ssel, "lsparseMatrix"), identical3(as.mat(sel.), as.mat(sel), as.mat(ssel)), identical3(!sel, !ssel, nsel), # ! is typically dense identical3(m[ sel], m[ ssel], as.matrix(m)[as.matrix( ssel)]), identical3(m[!sel], m[!ssel], as.matrix(m)[as.matrix(!ssel)]) ) ## more sparse Matrices -------------------------------------- m <- 1:800 set.seed(101) ; m[sample(800, 600)] <- 0 m <- Matrix(m, nrow = 40) mm <- as(m, "matrix") dimnames(mm) <- NULL ## << workaround: as(, "matrix") has NULL dimnames str(mC <- as(m, "dgCMatrix")) str(mT <- as(m, "dgTMatrix")) stopifnot(identical(mT, as(mC, "dgTMatrix")), identical(mC, as(mT, "dgCMatrix"))) mC[,1] mC[1:2,] mC[7, drop = FALSE] assert.EQ.mat(mC[1:2,], mm[1:2,]) ## *repeated* (aka 'duplicated') indices - did not work at all ... i <- rep(8:10,2) j <- c(2:4, 4:3) assert.EQ.mat(mC[i,], mm[i,]) assert.EQ.mat(mC[,j], mm[,j]) assert.EQ.mat(mC[i, 2:1], mm[i, 2:1]) assert.EQ.mat(mC[c(4,1,2:1), j], mm[c(4,1,2:1), j]) assert.EQ.mat(mC[i,j], mm[i,j]) set.seed(7) for(n in 1:50) { i <- sample(sample(nrow(mC), 7), 20, replace = TRUE) j <- sample(sample(ncol(mC), 6), 17, replace = TRUE) assert.EQ.mat(mC[i,j], mm[i,j]) } ##---- Symmetric indexing of symmetric Matrix ---------- m. <- mC; m.[, c(2, 7:12)] <- 0 validObject(S <- crossprod(add.simpleDimnames(m.) %% 100)) ss <- as(S, "matrix") T <- as(S, "TsparseMatrix") ## non-repeated indices: i <- c(7:5, 2:4);assert.EQ.mat(T[i,i], ss[i,i]) N <- nrow(T) set.seed(11) for(n in 1:50) { i <- sample(N, max(2, sample(N,1)), replace = FALSE) validObject(Tii <- T[i,i]) stopifnot(is(Tii, "dsTMatrix"), # remained symmetric Tsparse identical(t(Tii), t(T)[i,i])) assert.EQ.mat(Tii, ss[i,i]) } ## repeated ones ``the challenge'' (to do smartly): j <- c(4, 4, 9, 12, 9, 4, 17, 3, 18, 4, 12, 18, 4, 9) assert.EQ.mat(T[j,j], ss[j,j]) ## and another two sets (a, A) & (a., A.) : a <- matrix(0, 6,6) a[upper.tri(a)] <- (utr <- c(2, 0,-1, 0,0,5, 7,0,0,0, 0,0,-2,0,8)) ta <- t(a); ta[upper.tri(a)] <- utr; a <- t(ta) diag(a) <- c(0,3,0,4,6,0) A <- as(Matrix(a), "TsparseMatrix") A. <- A diag(A.) <- 10 * (1:6) a. <- as(A., "matrix") ## More testing {this was not working for a long time..} set.seed(1) for(n in 1:100) { i <- sample(1:nrow(A), 3+2*rpois(1, lam=3), replace=TRUE) Aii <- A[i,i] A.ii <- A.[i,i] stopifnot(class(Aii) == class(A), class(A.ii) == class(A.)) assert.EQ.mat(Aii , a [i,i]) assert.EQ.mat(A.ii, a.[i,i]) assert.EQ.mat(T[i,i], ss[i,i]) } stopifnot(all.equal(mC[,3], mm[,3]), identical(mC[ij], mm[ij])) assert.EQ.mat(mC[7, , drop=FALSE], mm[7, , drop=FALSE]) stopifnot(dim(mC[numeric(0), ]) == c(0,20), # used to give warnings dim(mC[, integer(0)]) == c(40,0), identical(mC[, integer(0)], mC[, FALSE]), identical(mC[7, drop = FALSE], mC[7,, drop = FALSE])) validObject(print(mT[,c(2,4)])) stopifnot(all.equal(mT[2,], mm[2,]), ## row or column indexing in combination with t() : identical(mT[2,], t(mT)[,2]), identical(mT[-2,], t(t(mT)[,-2])), identical(mT[c(2,5),], t(t(mT)[,c(2,5)])) ) assert.EQ.mat(mT[4,, drop = FALSE], mm[4,, drop = FALSE]) stopifnot(identical3(mm[,1], mC[,1], mT[,1]), identical3(mm[3,], mC[3,], mT[3,]), identical3(mT[2,3], mC[2,3], 0), identical(mT[], mT), identical4( mm[c(3,7), 2:4], as.mat( m[c(3,7), 2:4]), as.mat(mT[c(3,7), 2:4]), as.mat(mC[c(3,7), 2:4])) ) x.x <- crossprod(mC) stopifnot(class(x.x) == "dsCMatrix", class(x.x. <- round(x.x / 10000)) == "dsCMatrix", identical(x.x[cbind(2:6, 2:6)], diag(x.x [2:6, 2:6]))) head(x.x.) # Note the *non*-structural 0's printed as "0" tail(x.x., -3) # all but the first three lines lx.x <- as(x.x, "lsCMatrix") # FALSE only for "structural" 0 (l10 <- lx.x[1:10, 1:10])# "lsC" (l3 <- lx.x[1:3, ]) m.x <- as.mat(x.x) # as.mat() *drops* (NULL,NULL) dimnames stopifnot(class(l10) == "lsCMatrix", # symmetric indexing -> symmetric ! identical(as.mat(lx.x), m.x != 0), identical(as.logical(lx.x), as.logical(m.x)), identical(as.mat(l10), m.x[1:10, 1:10] != 0), identical(as.mat(l3 ), m.x[1:3, ] != 0) ) ##-- Sub*assignment* with repeated / duplicated index: A <- Matrix(0,4,3) ; A[c(1,2,1), 2] <- 1 ; A B <- A; B[c(1,2,1), 2] <- 1:3; B; B. <- B B.[3,] <- rbind(4:2) diag(B.) <- 10 * diag(B.) C <- B.; C[,2] <- C[,2]; C[1,] <- C[1,]; C[2:3,2:1] <- C[2:3,2:1] stopifnot(identical(unname(as.matrix(A)), local({a <- matrix(0,4,3); a[c(1,2,1), 2] <- 1 ; a})), identical(unname(as.matrix(B)), local({a <- matrix(0,4,3); a[c(1,2,1), 2] <- 1:3; a})), identical(C, drop0(B.))) ## used to fail n <- 5 ## or much larger sm <- new("dsTMatrix", i=as.integer(1),j=as.integer(1), Dim=as.integer(c(n,n)), x = 1) (cm <- as(sm, "CsparseMatrix")) sm[2,] stopifnot(sm[2,] == c(0:1, rep.int(0,ncol(sm)-2)), sm[2,] == cm[2,], sm[,3] == sm[3,], all(sm[,-(1:3)] == t(sm[-(1:3),])), # all() all(sm[,-(1:3)] == 0) ) m0 <- Diagonal(5) stopifnot(identical(m0[2,], m0[,2]), identical(m0[,1], c(1,0,0,0,0))) ### Diagonal -- Sparse: (m1 <- as(m0, "sparseMatrix")) # dtTMatrix (m2 <- as(m0, "CsparseMatrix")) # dtCMatrix (with an irrelevant warning) m1g <- as(m1, "generalMatrix") stopifnot(is(m1g, "dgTMatrix")) assert.EQ.mat(m2[1:3,], diag(5)[1:3,]) assert.EQ.mat(m2[,c(4,1)], diag(5)[,c(4,1)]) stopifnot(identical(m2[1:3,], as(m1[1:3,], "CsparseMatrix")), identical(Matrix:::uniqTsparse(m1[, c(4,2)]), Matrix:::uniqTsparse(as(m2[, c(4,2)], "TsparseMatrix"))) )## failed in 0.9975-11 (uTr <- new("dtTMatrix", Dim = c(3L,3L), diag="U")) uTr[1,] <- 0 assert.EQ.mat(uTr, cbind(0, rbind(0,diag(2)))) M <- m0; M[1,] <- 0 stopifnot(identical(M, Diagonal(x=c(0, rep(1,4))))) M <- m0; M[,3] <- 3 ; M ; stopifnot(is(M, "sparseMatrix"), M[,3] == 3) validObject(M) M <- m0; M[1:3, 3] <- 0 ;M T <- m0; T[1:3, 3] <- 10 stopifnot(identical(M, Diagonal(x=c(1,1, 0, 1,1))), is(T, "triangularMatrix"), identical(T[,3], c(10,10,10,0,0))) M <- m1; M[1,] <- 0 ; M ; assert.EQ.mat(M, diag(c(0,rep(1,4))), tol=0) M <- m1; M[,3] <- 3 ; stopifnot(is(M,"sparseMatrix"), M[,3] == 3) validObject(M) M <- m1; M[1:3, 3] <- 0 ;M assert.EQ.mat(M, diag(c(1,1, 0, 1,1)), tol=0) T <- m1; T[1:3, 3] <- 10; validObject(T) stopifnot(is(T, "dtTMatrix"), identical(T[,3], c(10,10,10,0,0))) M <- m2; M[1,] <- 0 ; M ; assert.EQ.mat(M, diag(c(0,rep(1,4))), tol=0) M <- m2; M[,3] <- 3 ; stopifnot(is(M,"sparseMatrix"), M[,3] == 3) validObject(M) M <- m2; M[1:3, 3] <- 0 ;M assert.EQ.mat(M, diag(c(1,1, 0, 1,1)), tol=0) T <- m2; T[1:3, 3] <- 10; validObject(T) stopifnot(is(T, "dtCMatrix"), identical(T[,3], c(10,10,10,0,0))) ## "Vector indices" ------------------- D <- Diagonal(6) M <- as(D,"dgeMatrix") m <- as(D,"matrix") s <- as(D,"TsparseMatrix") S <- as(s,"CsparseMatrix") i <- c(3,1,6); v <- c(10,15,20) ## (logical,value) which both are recycled: L <- c(TRUE, rep(FALSE,8)) ; z <- c(50,99) ## vector subassignment, both with integer & logical ## these now work correctly {though not very efficiently; hence warnings} m[i] <- v # the role model: only first column is affected M[i] <- v; assert.EQ.mat(M,m) # dge D[i] <- v; assert.EQ.mat(D,m) # ddi -> dtT -> dgT s[i] <- v; assert.EQ.mat(s,m) # dtT -> dgT S[i] <- v; assert.EQ.mat(S,m); S # dtC -> dtT -> dgT -> dgC ## logical m[L] <- z M[L] <- z; assert.EQ.mat(M,m) D[L] <- z; assert.EQ.mat(D,m) s[L] <- z; assert.EQ.mat(s,m) S[L] <- z; assert.EQ.mat(S,m) ; S ## indexing [i] vs [i,] --- now ok stopifnot(identical4(m[i], M[i], D[i], s[i]), identical(s[i],S[i])) stopifnot(identical4(m[L], M[L], D[L], s[L]), identical(s[L],S[L])) assert.EQ.mat(D[i,], m[i,]) assert.EQ.mat(M[i,], m[i,]) assert.EQ.mat(s[i,], m[i,]) assert.EQ.mat(S[i,], m[i,]) assert.EQ.mat(D[,i], m[,i]) assert.EQ.mat(M[,i], m[,i]) assert.EQ.mat(s[,i], m[,i]) assert.EQ.mat(S[,i], m[,i]) ## --- negative indices ---------- mc <- mC[1:5, 1:7] mt <- mT[1:5, 1:7] ## sub matrix assert.EQ.mat(mC[1:2, 0:3], mm[1:2, 0:3]) # test 0-index stopifnot(identical(mc[-(3:5), 0:2], mC[1:2, 0:2]), identical(mt[-(3:5), 0:2], mT[1:2, 0:2]), identical(mC[2:3, 4], mm[2:3, 4])) assert.EQ.mat(mC[1:2,], mm[1:2,]) ## sub vector stopifnot(identical4(mc[-(1:4), ], mC[5, 1:7], mt[-(1:4), ], mT[5, 1:7])) stopifnot(identical4(mc[-(1:4), -(2:4)], mC[5, c(1,5:7)], mt[-(1:4), -(2:4)], mT[5, c(1,5:7)])) ## mixing of negative and positive must give error assertError(mT[-1:1,]) ## Sub *Assignment* ---- now works (partially): mt0 <- mt mt[1, 4] <- -99 mt[2:3, 1:6] <- 0 mt m2 <- mt+mt m2[1,4] <- -200 m2[c(1,3), c(5:6,2)] <- 1:6 stopifnot(m2[1,4] == -200, as.vector(m2[c(1,3), c(5:6,2)]) == 1:6) mt[,3] <- 30 mt[2:3,] <- 250 mt[1:5 %% 2 == 1, 3] <- 0 mt[3:1, 1:7 > 5] <- 0 mt tt <- as(mt,"matrix") ii <- c(0,2,5) jj <- c(2:3,5) tt[ii, jj] <- 1:6 # 0 is just "dropped" mt[ii, jj] <- 1:6 assert.EQ.mat(mt, tt) mt[1:5, 2:6] as((mt0 - mt)[1:5,], "dsparseMatrix")# [1,5] and lines 2:3 mt[c(2,4), ] <- 0; stopifnot(as(mt[c(2,4), ],"matrix") == 0) mt[2:3, 4:7] <- 33 validObject(mt) mt mc[1,4] <- -99 ; stopifnot(mc[1,4] == -99) mc[1,4] <- 00 ; stopifnot(mc[1,4] == 00) mc[1,4] <- -99 ; stopifnot(mc[1,4] == -99) mc[1:2,4:3] <- 4:1; stopifnot(as.matrix(mc[1:2,4:3]) == 4:1) mc[-1, 3] <- -2:1 # 0 should not be entered; 'value' recycled mt[-1, 3] <- -2:1 stopifnot(mc@x != 0, mt@x != 0, mc[-1,3] == -2:1, mt[-1,3] == -2:1) ## failed earlier mc0 <- mc mt0 <- as(mc0, "TsparseMatrix") m0 <- as(mc0, "matrix") set.seed(1) for(i in 1:50) { mc <- mc0; mt <- mt0 ; m <- m0 ev <- 1:5 %% 2 == round(runif(1))# 0 or 1 j <- sample(ncol(mc), 1 + round(runif(1))) nv <- rpois(sum(ev) * length(j), lambda = 1) mc[ev, j] <- nv m[ev, j] <- nv mt[ev, j] <- nv if(i %% 10 == 1) print(mc[ev,j, drop = FALSE]) stopifnot(as.vector(mc[ev, j]) == nv, ## failed earlier... as.vector(mt[ev, j]) == nv) validObject(mc) ; assert.EQ.mat(mc, m) validObject(mt) ; assert.EQ.mat(mt, m) } mc # no longer has non-structural zeros mc[ii, jj] <- 1:6 mc[c(2,5), c(3,5)] <- 3.2 validObject(mc) m. <- mc mc[4,] <- 0 mc S <- as(Diagonal(5),"sparseMatrix") H <- Hilbert(9) Hc <- as(round(H, 3), "dsCMatrix")# a sparse matrix with no 0 ... (trH <- tril(Hc[1:5, 1:5])) stopifnot(is(trH, "triangularMatrix"), trH@uplo == "L", is(S, "triangularMatrix")) ## triangular assignment ## the slick (but inefficient in case of sparse!) way to assign sub-diagonals: ## equivalent to tmp <- `diag<-`(S[,-1], -2:1); S[,-1] <- tmp ## which dispatches to (x="TsparseMatrix", i="missing",j="index", value="replValue") diag(S[,-1]) <- -2:1 # used to give a wrong warning S <- as(S,"triangularMatrix") assert.EQ.mat(S, local({s <- diag(5); diag(s[,-1]) <- -2:1; s})) trH[c(1:2,4), c(2:3,5)] <- 0 # gave an *error* upto Jan.2008 trH[ lower.tri(trH) ] <- 0 # ditto, because of callNextMethod() m <- Matrix(0+1:28, nrow = 4) m[-3,c(2,4:5,7)] <- m[ 3, 1:4] <- m[1:3, 6] <- 0 mT <- as(m, "dgTMatrix") stopifnot(identical(mT[lower.tri(mT)], m [lower.tri(m) ])) lM <- upper.tri(mT, diag=TRUE) mT[lM] <- 0 m[lM] <- 0 assert.EQ.mat(mT, as(m,"matrix")) mT[lM] <- -1:0 m[lM] <- -1:0 assert.EQ.mat(mT, as(m,"matrix")) (mT <- drop0(mT)) i <- c(1:2, 4, 6:7); j <- c(2:4,6) H[i,j] <- 0 (H. <- round(as(H, "sparseMatrix"), 3)[ , 2:7]) Hc. <- Hc Hc.[i,j] <- 0 ## now "works", but setting "non-structural" 0s stopifnot(as.matrix(Hc.[i,j]) == 0) Hc.[, 1:6] ## an example that failed for a long time sy3 <- new("dsyMatrix", Dim = as.integer(c(2, 2)), x = c(14, -1, 2, -7)) validObject(dm <- kronecker(Diagonal(2), sy3))# now sparse with new kronecker dm <- Matrix(as.matrix(dm))# -> "dsyMatrix" (s2 <- as(dm, "sparseMatrix")) validObject(st <- as(s2, "TsparseMatrix")) stopifnot(is(s2, "symmetricMatrix"), is(st, "symmetricMatrix")) validObject(s.32 <- st[1:3,1:2]) ## 3 x 2 - and *not* dsTMatrix validObject(s2.32 <- s2[1:3,1:2]) I <- c(1,4:3) stopifnot(is(s2.32, "generalMatrix"), is(s.32, "generalMatrix"), identical(as.mat(s.32), as.mat(s2.32)), identical3(dm[1:3,-1], asD(s2[1:3,-1]), asD(st[1:3,-1])), identical4(2, dm[4,3], s2[4,3], st[4,3]), identical3(diag(dm), diag(s2), diag(st)), is((cI <- s2[I,I]), "dsCMatrix"), is((tI <- st[I,I]), "dsTMatrix"), identical4(as.mat(dm)[I,I], as.mat(dm[I,I]), as.mat(tI), as.mat(cI)) ) ## now sub-assign and check for consistency ## symmetric subassign should keep symmetry st[I,I] <- 0; validObject(st); stopifnot(is(st,"symmetricMatrix")) s2[I,I] <- 0; validObject(s2); stopifnot(is(s2,"symmetricMatrix")) ## m <- as.mat(st) m[2:1,2:1] <- 4:1 st[2:1,2:1] <- 4:1 s2[2:1,2:1] <- 4:1 stopifnot(identical(m, as.mat(st)), 1:4 == as.vector(s2[1:2,1:2]), identical(m, as.mat(s2))) ## now a slightly different situation for 's2' (had bug) s2 <- as(dm, "sparseMatrix") s2[I,I] <- 0; diag(s2)[2:3] <- -(1:2) stopifnot(is(s2,"symmetricMatrix"), diag(s2) == c(0:-2,0)) t2 <- as(s2, "TsparseMatrix") m <- as.mat(s2) s2[2:1,2:1] <- 4:1 t2[2:1,2:1] <- 4:1 m[2:1,2:1] <- 4:1 assert.EQ.mat(t2, m) assert.EQ.mat(s2, m) ## and the same (for a different s2 !) s2[2:1,2:1] <- 4:1 t2[2:1,2:1] <- 4:1 assert.EQ.mat(t2, m)# ok assert.EQ.mat(s2, m)# failed in 0.9975-8 ## m[cbind(i,j)] <- value: m.[ cbind(3:5, 1:3) ] <- 1:3 stopifnot(m.[3,1] == 1, m.[4,2] == 2) x.x[ cbind(2:6, 2:6)] <- 12:16 validObject(x.x) stopifnot(class(x.x) == "dsCMatrix", 12:16 == as.mat(x.x)[cbind(2:6, 2:6)]) (ne1 <- (mc - m.) != 0) stopifnot(identical(ne1, 0 != abs(mc - m.))) (ge <- m. >= mc) # contains "=" -> result is dense ne. <- mc != m. # was wrong (+ warning) stopifnot(identical(!(m. < mc), m. >= mc), identical(m. < mc, as(!ge, "sparseMatrix")), identical(ne., drop0(ne1))) (M3 <- Matrix(upper.tri(matrix(, 3, 3)))) # ltC; indexing used to fail T3 <- as(M3, "TsparseMatrix") stopifnot(identical(drop(M3), M3), identical4(drop(M3[,2, drop = FALSE]), M3[,2, drop = TRUE], drop(T3[,2, drop = FALSE]), T3[,2, drop = TRUE]), is(T3, "triangularMatrix"), !is(T3[,2, drop=FALSE], "triangularMatrix") ) M <- Diagonal(4); M[1,2] <- 2 M. <- as(M, "CsparseMatrix") (R <- as(M., "RsparseMatrix")) stopifnot(is(M, "triangularMatrix"), is(M.,"triangularMatrix"), is(R, "triangularMatrix")) stopifnot(dim(M[2:3, FALSE]) == c(2,0), dim(R[2:3, FALSE]) == c(2,0), identical(M [2:3,TRUE], M [2:3,]), identical(M.[2:3,TRUE], M.[2:3,]), identical(R [2:3,TRUE], R [2:3,]), dim(R[FALSE, FALSE]) == c(0,0)) cat('Time elapsed: ', proc.time(),'\n') # for ``statistical reasons''