matrixStats.benchmarks
colLogSumExps() and rowLogSumExps() benchmarks
This report benchmark the performance of colLogSumExps() and rowLogSumExps() against alternative methods.
Alternative methods
- apply() + matrixStats::logSumExp()
- apply() + logSumExp0()
where
> logSumExp0 <- function(lx, ...) {
+ iMax <- which.max(lx)
+ log1p(sum(exp(lx[-iMax] - lx[iMax]))) + lx[iMax]
+ }
Data
> rmatrix <- function(nrow, ncol, mode = c("logical", "double", "integer", "index"), range = c(-100,
+ +100), na_prob = 0) {
+ mode <- match.arg(mode)
+ n <- nrow * ncol
+ if (mode == "logical") {
+ x <- sample(c(FALSE, TRUE), size = n, replace = TRUE)
+ } else if (mode == "index") {
+ x <- seq_len(n)
+ mode <- "integer"
+ } else {
+ x <- runif(n, min = range[1], max = range[2])
+ }
+ storage.mode(x) <- mode
+ if (na_prob > 0)
+ x[sample(n, size = na_prob * n)] <- NA
+ dim(x) <- c(nrow, ncol)
+ x
+ }
> rmatrices <- function(scale = 10, seed = 1, ...) {
+ set.seed(seed)
+ data <- list()
+ data[[1]] <- rmatrix(nrow = scale * 1, ncol = scale * 1, ...)
+ data[[2]] <- rmatrix(nrow = scale * 10, ncol = scale * 10, ...)
+ data[[3]] <- rmatrix(nrow = scale * 100, ncol = scale * 1, ...)
+ data[[4]] <- t(data[[3]])
+ data[[5]] <- rmatrix(nrow = scale * 10, ncol = scale * 100, ...)
+ data[[6]] <- t(data[[5]])
+ names(data) <- sapply(data, FUN = function(x) paste(dim(x), collapse = "x"))
+ data
+ }
> data <- rmatrices(mode = "double")
Results
10x10 matrix
> X <- data[["10x10"]]
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5242264 280.0 7916910 422.9 7916910 422.9
Vcells 10188779 77.8 33191153 253.3 53339345 407.0
> colStats <- microbenchmark(colLogSumExps = colLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 2L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 2L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
> X <- t(X)
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5228192 279.3 7916910 422.9 7916910 422.9
Vcells 10142210 77.4 33191153 253.3 53339345 407.0
> rowStats <- microbenchmark(rowLogSumExps = rowLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 1L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 1L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
Table: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 10x10 data. The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 0.008133 | 0.0090895 | 0.0104243 | 0.0104435 | 0.0109565 | 0.032976 |
| 2 | apply+logSumExp | 0.047248 | 0.0507385 | 0.0535950 | 0.0527455 | 0.0533990 | 0.155811 |
| 3 | apply+logSumExp0 | 0.064913 | 0.0685595 | 0.0716377 | 0.0705830 | 0.0714225 | 0.174637 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 |
| 2 | apply+logSumExp | 5.809418 | 5.582100 | 5.141343 | 5.050558 | 4.873728 | 4.724982 |
| 3 | apply+logSumExp0 | 7.981434 | 7.542714 | 6.872169 | 6.758558 | 6.518733 | 5.295882 |
Table: Benchmarking of rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 10x10 data (transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 0.007400 | 0.0081525 | 0.0094867 | 0.0093865 | 0.0099785 | 0.028806 |
| 2 | apply+logSumExp | 0.047769 | 0.0521430 | 0.0536802 | 0.0528420 | 0.0533980 | 0.085639 |
| 3 | apply+logSumExp0 | 0.065200 | 0.0696110 | 0.0724195 | 0.0713380 | 0.0721935 | 0.155864 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 |
| 2 | apply+logSumExp | 6.455270 | 6.395952 | 5.658454 | 5.629574 | 5.351305 | 2.972957 |
| 3 | apply+logSumExp0 | 8.810811 | 8.538608 | 7.633780 | 7.600064 | 7.234905 | 5.410817 |
Figure: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 10x10 data as well as rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on the same data transposed. Outliers are displayed as crosses. Times are in milliseconds.
Table: Benchmarking of colLogSumExps() and rowLogSumExps() on 10x10 data (original and transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 2 | rowLogSumExps | 7.400 | 8.1525 | 9.48672 | 9.3865 | 9.9785 | 28.806 |
| 1 | colLogSumExps | 8.133 | 9.0895 | 10.42432 | 10.4435 | 10.9565 | 32.976 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 2 | rowLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 |
| 1 | colLogSumExps | 1.099054 | 1.114934 | 1.098833 | 1.112608 | 1.098011 | 1.144761 |
Figure: Benchmarking of colLogSumExps() and rowLogSumExps() on 10x10 data (original and transposed). Outliers are displayed as crosses. Times are in milliseconds.
100x100 matrix
> X <- data[["100x100"]]
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5226752 279.2 7916910 422.9 7916910 422.9
Vcells 9758795 74.5 33191153 253.3 53339345 407.0
> colStats <- microbenchmark(colLogSumExps = colLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 2L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 2L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
> X <- t(X)
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5226746 279.2 7916910 422.9 7916910 422.9
Vcells 9768838 74.6 33191153 253.3 53339345 407.0
> rowStats <- microbenchmark(rowLogSumExps = rowLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 1L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 1L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
Table: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 100x100 data. The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 0.147286 | 0.1533570 | 0.1748630 | 0.1615145 | 0.1881190 | 0.318889 |
| 2 | apply+logSumExp | 0.374943 | 0.3857225 | 0.4386438 | 0.4127325 | 0.4663460 | 0.766932 |
| 3 | apply+logSumExp0 | 0.532599 | 0.5451135 | 0.6201473 | 0.5709585 | 0.6800395 | 1.043200 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 |
| 2 | apply+logSumExp | 2.545680 | 2.515193 | 2.508500 | 2.555390 | 2.478995 | 2.405012 |
| 3 | apply+logSumExp0 | 3.616087 | 3.554539 | 3.546476 | 3.535029 | 3.614943 | 3.271358 |
Table: Benchmarking of rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 100x100 data (transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 0.151541 | 0.1609810 | 0.1803268 | 0.1699135 | 0.192550 | 0.266127 |
| 2 | apply+logSumExp | 0.375183 | 0.3909565 | 0.4443629 | 0.4233270 | 0.470347 | 0.679589 |
| 3 | apply+logSumExp0 | 0.534134 | 0.5546200 | 0.6243859 | 0.5808370 | 0.671827 | 1.073041 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 |
| 2 | apply+logSumExp | 2.475785 | 2.428588 | 2.464208 | 2.491427 | 2.442727 | 2.553627 |
| 3 | apply+logSumExp0 | 3.524683 | 3.445251 | 3.462523 | 3.418428 | 3.489104 | 4.032064 |
Figure: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 100x100 data as well as rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on the same data transposed. Outliers are displayed as crosses. Times are in milliseconds.
Table: Benchmarking of colLogSumExps() and rowLogSumExps() on 100x100 data (original and transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 147.286 | 153.357 | 174.8630 | 161.5145 | 188.119 | 318.889 |
| 2 | rowLogSumExps | 151.541 | 160.981 | 180.3268 | 169.9135 | 192.550 | 266.127 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.0000000 |
| 2 | rowLogSumExps | 1.028889 | 1.049714 | 1.031247 | 1.052002 | 1.023554 | 0.8345443 |
Figure: Benchmarking of colLogSumExps() and rowLogSumExps() on 100x100 data (original and transposed). Outliers are displayed as crosses. Times are in milliseconds.
1000x10 matrix
> X <- data[["1000x10"]]
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5227507 279.2 7916910 422.9 7916910 422.9
Vcells 9762590 74.5 33191153 253.3 53339345 407.0
> colStats <- microbenchmark(colLogSumExps = colLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 2L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 2L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
> X <- t(X)
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5227495 279.2 7916910 422.9 7916910 422.9
Vcells 9772623 74.6 33191153 253.3 53339345 407.0
> rowStats <- microbenchmark(rowLogSumExps = rowLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 1L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 1L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
Table: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 1000x10 data. The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 0.148651 | 0.1647045 | 0.1906819 | 0.1847195 | 0.2085615 | 0.350757 |
| 2 | apply+logSumExp | 0.227003 | 0.2507095 | 0.2880074 | 0.2765740 | 0.3024020 | 0.486887 |
| 3 | apply+logSumExp0 | 0.269485 | 0.3098875 | 0.3494949 | 0.3436425 | 0.3780010 | 0.480715 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 |
| 2 | apply+logSumExp | 1.527087 | 1.522178 | 1.510408 | 1.497265 | 1.449942 | 1.388103 |
| 3 | apply+logSumExp0 | 1.812870 | 1.881476 | 1.832869 | 1.860348 | 1.812420 | 1.370507 |
Table: Benchmarking of rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 1000x10 data (transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 0.148117 | 0.1623880 | 0.1906410 | 0.1862295 | 0.2183835 | 0.266496 |
| 2 | apply+logSumExp | 0.216779 | 0.2381875 | 0.2746292 | 0.2602685 | 0.3046930 | 0.471987 |
| 3 | apply+logSumExp0 | 0.261497 | 0.2902070 | 0.3273246 | 0.3186795 | 0.3461150 | 0.478835 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 |
| 2 | apply+logSumExp | 1.463566 | 1.466780 | 1.440557 | 1.397569 | 1.395220 | 1.771085 |
| 3 | apply+logSumExp0 | 1.765476 | 1.787121 | 1.716968 | 1.711219 | 1.584895 | 1.796781 |
Figure: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 1000x10 data as well as rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on the same data transposed. Outliers are displayed as crosses. Times are in milliseconds.
Table: Benchmarking of colLogSumExps() and rowLogSumExps() on 1000x10 data (original and transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 148.651 | 164.7045 | 190.6819 | 184.7195 | 208.5615 | 350.757 |
| 2 | rowLogSumExps | 148.117 | 162.3880 | 190.6410 | 186.2295 | 218.3835 | 266.496 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.0000000 | 1.0000000 | 1.0000000 | 1.000000 | 1.000000 | 1.0000000 |
| 2 | rowLogSumExps | 0.9964077 | 0.9859354 | 0.9997857 | 1.008175 | 1.047094 | 0.7597739 |
Figure: Benchmarking of colLogSumExps() and rowLogSumExps() on 1000x10 data (original and transposed). Outliers are displayed as crosses. Times are in milliseconds.
10x1000 matrix
> X <- data[["10x1000"]]
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5227714 279.2 7916910 422.9 7916910 422.9
Vcells 9763394 74.5 33191153 253.3 53339345 407.0
> colStats <- microbenchmark(colLogSumExps = colLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 2L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 2L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
> X <- t(X)
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5227708 279.2 7916910 422.9 7916910 422.9
Vcells 9773437 74.6 33191153 253.3 53339345 407.0
> rowStats <- microbenchmark(rowLogSumExps = rowLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 1L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 1L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
Table: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 10x1000 data. The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 0.182868 | 0.198522 | 0.2186889 | 0.206460 | 0.2248295 | 0.425178 |
| 2 | apply+logSumExp | 1.811853 | 1.978557 | 2.1914504 | 2.065537 | 2.1991665 | 4.675994 |
| 3 | apply+logSumExp0 | 3.031848 | 3.310961 | 3.8202764 | 3.474716 | 3.6859950 | 12.717593 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.000000 | 1.000000 | 1.00000 | 1.00000 | 1.000000 | 1.00000 |
| 2 | apply+logSumExp | 9.907983 | 9.966439 | 10.02086 | 10.00454 | 9.781486 | 10.99773 |
| 3 | apply+logSumExp0 | 16.579434 | 16.678058 | 17.46900 | 16.82997 | 16.394624 | 29.91122 |
Table: Benchmarking of rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 10x1000 data (transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 0.177854 | 0.2003505 | 0.2230796 | 0.2116575 | 0.2355875 | 0.378667 |
| 2 | apply+logSumExp | 1.798603 | 1.9887350 | 2.1438652 | 2.0944100 | 2.2260755 | 3.235006 |
| 3 | apply+logSumExp0 | 3.021550 | 3.3278055 | 3.5985985 | 3.4443350 | 3.6162465 | 8.963276 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 1.00000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 |
| 2 | apply+logSumExp | 10.11281 | 9.926279 | 9.610316 | 9.895279 | 9.449039 | 8.543142 |
| 3 | apply+logSumExp0 | 16.98893 | 16.609919 | 16.131456 | 16.273154 | 15.349908 | 23.670602 |
Figure: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 10x1000 data as well as rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on the same data transposed. Outliers are displayed as crosses. Times are in milliseconds.
Table: Benchmarking of colLogSumExps() and rowLogSumExps() on 10x1000 data (original and transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 182.868 | 198.5220 | 218.6889 | 206.4600 | 224.8295 | 425.178 |
| 2 | rowLogSumExps | 177.854 | 200.3505 | 223.0796 | 211.6575 | 235.5875 | 378.667 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.0000000 | 1.000000 | 1.000000 | 1.000000 | 1.00000 | 1.0000000 |
| 2 | rowLogSumExps | 0.9725813 | 1.009211 | 1.020077 | 1.025174 | 1.04785 | 0.8906082 |
Figure: Benchmarking of colLogSumExps() and rowLogSumExps() on 10x1000 data (original and transposed). Outliers are displayed as crosses. Times are in milliseconds.
100x1000 matrix
> X <- data[["100x1000"]]
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5227930 279.3 7916910 422.9 7916910 422.9
Vcells 9763989 74.5 33191153 253.3 53339345 407.0
> colStats <- microbenchmark(colLogSumExps = colLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 2L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 2L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
> X <- t(X)
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5227912 279.3 7916910 422.9 7916910 422.9
Vcells 9864012 75.3 33191153 253.3 53339345 407.0
> rowStats <- microbenchmark(rowLogSumExps = rowLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 1L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 1L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
Table: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 100x1000 data. The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.537931 | 1.665246 | 1.759681 | 1.704182 | 1.837969 | 2.555003 |
| 2 | apply+logSumExp | 3.701825 | 4.101192 | 4.849556 | 4.249950 | 4.569222 | 26.022281 |
| 3 | apply+logSumExp0 | 5.369943 | 5.892519 | 6.522110 | 6.149206 | 6.676341 | 28.776985 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.00000 |
| 2 | apply+logSumExp | 2.407016 | 2.462814 | 2.755929 | 2.493836 | 2.486017 | 10.18483 |
| 3 | apply+logSumExp0 | 3.491667 | 3.538527 | 3.706415 | 3.608304 | 3.632455 | 11.26299 |
Table: Benchmarking of rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 100x1000 data (transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 1.580767 | 1.732992 | 2.018422 | 1.765429 | 1.823624 | 21.395092 |
| 2 | apply+logSumExp | 3.719148 | 4.072647 | 4.282412 | 4.166019 | 4.295800 | 5.740941 |
| 3 | apply+logSumExp0 | 5.225207 | 5.850327 | 6.585593 | 5.939724 | 6.127233 | 29.342143 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.0000000 |
| 2 | apply+logSumExp | 2.352749 | 2.350067 | 2.121663 | 2.359777 | 2.355639 | 0.2683298 |
| 3 | apply+logSumExp0 | 3.305488 | 3.375853 | 3.262744 | 3.364465 | 3.359921 | 1.3714427 |
Figure: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 100x1000 data as well as rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on the same data transposed. Outliers are displayed as crosses. Times are in milliseconds.
Table: Benchmarking of colLogSumExps() and rowLogSumExps() on 100x1000 data (original and transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.537931 | 1.665246 | 1.759681 | 1.704182 | 1.837969 | 2.555003 |
| 2 | rowLogSumExps | 1.580767 | 1.732992 | 2.018422 | 1.765429 | 1.823624 | 21.395092 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.0000000 | 1.000000 |
| 2 | rowLogSumExps | 1.027853 | 1.040682 | 1.147038 | 1.035939 | 0.9921952 | 8.373803 |
Figure: Benchmarking of colLogSumExps() and rowLogSumExps() on 100x1000 data (original and transposed). Outliers are displayed as crosses. Times are in milliseconds.
1000x100 matrix
> X <- data[["1000x100"]]
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5228130 279.3 7916910 422.9 7916910 422.9
Vcells 9764643 74.5 33191153 253.3 53339345 407.0
> colStats <- microbenchmark(colLogSumExps = colLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 2L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 2L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
> X <- t(X)
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 5228118 279.3 7916910 422.9 7916910 422.9
Vcells 9864676 75.3 33191153 253.3 53339345 407.0
> rowStats <- microbenchmark(rowLogSumExps = rowLogSumExps(X, na.rm = FALSE), `apply+logSumExp` = apply(X,
+ MARGIN = 1L, FUN = logSumExp, na.rm = FALSE), `apply+logSumExp0` = apply(X, MARGIN = 1L, FUN = logSumExp0,
+ na.rm = FALSE), unit = "ms")
Table: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 1000x100 data. The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.461910 | 1.665638 | 1.874952 | 1.758566 | 1.838413 | 9.37362 |
| 2 | apply+logSumExp | 2.183029 | 2.406405 | 2.725101 | 2.592052 | 2.748403 | 10.75560 |
| 3 | apply+logSumExp0 | 2.546085 | 2.931929 | 3.311470 | 3.118014 | 3.495567 | 10.58627 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | colLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 |
| 2 | apply+logSumExp | 1.493272 | 1.444735 | 1.453425 | 1.473958 | 1.494986 | 1.147433 |
| 3 | apply+logSumExp0 | 1.741615 | 1.760244 | 1.766163 | 1.773044 | 1.901404 | 1.129369 |
Table: Benchmarking of rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 1000x100 data (transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 1.475570 | 1.718748 | 1.871399 | 1.757151 | 1.810914 | 9.543469 |
| 2 | apply+logSumExp | 2.189496 | 2.400485 | 2.563202 | 2.460184 | 2.625558 | 5.026601 |
| 3 | apply+logSumExp0 | 2.670531 | 2.930343 | 3.328314 | 2.997537 | 3.231213 | 10.639849 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 1 | rowLogSumExps | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.000000 | 1.0000000 |
| 2 | apply+logSumExp | 1.483831 | 1.396648 | 1.369671 | 1.400098 | 1.449852 | 0.5267059 |
| 3 | apply+logSumExp0 | 1.809830 | 1.704929 | 1.778517 | 1.705907 | 1.784299 | 1.1148828 |
Figure: Benchmarking of colLogSumExps(), apply+logSumExp() and apply+logSumExp0() on 1000x100 data as well as rowLogSumExps(), apply+logSumExp() and apply+logSumExp0() on the same data transposed. Outliers are displayed as crosses. Times are in milliseconds.
Table: Benchmarking of colLogSumExps() and rowLogSumExps() on 1000x100 data (original and transposed). The top panel shows times in milliseconds and the bottom panel shows relative times.
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 2 | rowLogSumExps | 1.47557 | 1.718748 | 1.871399 | 1.757151 | 1.810914 | 9.543469 |
| 1 | colLogSumExps | 1.46191 | 1.665638 | 1.874952 | 1.758566 | 1.838413 | 9.373620 |
| expr | min | lq | mean | median | uq | max | |
|---|---|---|---|---|---|---|---|
| 2 | rowLogSumExps | 1.0000000 | 1.0000000 | 1.000000 | 1.000000 | 1.000000 | 1.0000000 |
| 1 | colLogSumExps | 0.9907426 | 0.9690996 | 1.001898 | 1.000805 | 1.015185 | 0.9822026 |
Figure: Benchmarking of colLogSumExps() and rowLogSumExps() on 1000x100 data (original and transposed). Outliers are displayed as crosses. Times are in milliseconds.
Appendix
Session information
R version 4.1.1 Patched (2021-08-10 r80727)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 18.04.5 LTS
Matrix products: default
BLAS: /home/hb/software/R-devel/R-4-1-branch/lib/R/lib/libRblas.so
LAPACK: /home/hb/software/R-devel/R-4-1-branch/lib/R/lib/libRlapack.so
locale:
[1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
[3] LC_TIME=en_US.UTF-8 LC_COLLATE=en_US.UTF-8
[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
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] microbenchmark_1.4-7 matrixStats_0.60.0 ggplot2_3.3.5
[4] knitr_1.33 R.devices_2.17.0 R.utils_2.10.1
[7] R.oo_1.24.0 R.methodsS3_1.8.1-9001 history_0.0.1-9000
loaded via a namespace (and not attached):
[1] Biobase_2.52.0 httr_1.4.2 splines_4.1.1
[4] bit64_4.0.5 network_1.17.1 assertthat_0.2.1
[7] highr_0.9 stats4_4.1.1 blob_1.2.2
[10] GenomeInfoDbData_1.2.6 robustbase_0.93-8 pillar_1.6.2
[13] RSQLite_2.2.8 lattice_0.20-44 glue_1.4.2
[16] digest_0.6.27 XVector_0.32.0 colorspace_2.0-2
[19] Matrix_1.3-4 XML_3.99-0.7 pkgconfig_2.0.3
[22] zlibbioc_1.38.0 genefilter_1.74.0 purrr_0.3.4
[25] ergm_4.1.2 xtable_1.8-4 scales_1.1.1
[28] tibble_3.1.4 annotate_1.70.0 KEGGREST_1.32.0
[31] farver_2.1.0 generics_0.1.0 IRanges_2.26.0
[34] ellipsis_0.3.2 cachem_1.0.6 withr_2.4.2
[37] BiocGenerics_0.38.0 mime_0.11 survival_3.2-13
[40] magrittr_2.0.1 crayon_1.4.1 statnet.common_4.5.0
[43] memoise_2.0.0 laeken_0.5.1 fansi_0.5.0
[46] R.cache_0.15.0 MASS_7.3-54 R.rsp_0.44.0
[49] progressr_0.8.0 tools_4.1.1 lifecycle_1.0.0
[52] S4Vectors_0.30.0 trust_0.1-8 munsell_0.5.0
[55] tabby_0.0.1-9001 AnnotationDbi_1.54.1 Biostrings_2.60.2
[58] compiler_4.1.1 GenomeInfoDb_1.28.1 rlang_0.4.11
[61] grid_4.1.1 RCurl_1.98-1.4 cwhmisc_6.6
[64] rstudioapi_0.13 rappdirs_0.3.3 startup_0.15.0
[67] labeling_0.4.2 bitops_1.0-7 base64enc_0.1-3
[70] boot_1.3-28 gtable_0.3.0 DBI_1.1.1
[73] markdown_1.1 R6_2.5.1 lpSolveAPI_5.5.2.0-17.7
[76] rle_0.9.2 dplyr_1.0.7 fastmap_1.1.0
[79] bit_4.0.4 utf8_1.2.2 parallel_4.1.1
[82] Rcpp_1.0.7 vctrs_0.3.8 png_0.1-7
[85] DEoptimR_1.0-9 tidyselect_1.1.1 xfun_0.25
[88] coda_0.19-4
Total processing time was 17.62 secs.
Reproducibility
To reproduce this report, do:
html <- matrixStats:::benchmark('colLogSumExps')
Copyright Henrik Bengtsson. Last updated on 2021-08-25 22:16:45 (+0200 UTC). Powered by RSP.