1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
|
<!-- README.md is generated from README.Rmd. Please edit that file -->
# DelayedMatrixStats
[](https://travis-ci.org/PeteHaitch/DelayedMatrixStats)
[](https://codecov.io/gh/PeteHaitch/DelayedMatrixStats)
**DelayedMatrixStats** is a port of the
[**matrixStats**](https://CRAN.R-project.org/package=matrixStats) API to
work with *DelayedMatrix* objects from the
[**DelayedArray**](http://bioconductor.org/packages/DelayedArray/)
package.
For a *DelayedMatrix*, `x`, the simplest way to apply a function, `f()`,
from **matrixStats** is`matrixStats::f(as.matrix(x))`. However, this
“*realizes*” `x` in memory as a *base::matrix*, which typically
defeats the entire purpose of using a *DelayedMatrix* for storing the
data.
The **DelayedArray** package already implements a clever strategy called
“block-processing” for certain common “matrix stats” operations (e.g.
`colSums()`, `rowSums()`). This is a good start, but not all of the
**matrixStats** API is currently supported. Furthermore, certain
operations can be optimized with additional information about `x`. I’ll
refer to these “seed-aware” implementations.
## Installation
You can install **DelayedMatrixStats** from Bioconductor with:
``` r
if (!requireNamespace("BiocManager", quietly = TRUE))
install.packages("BiocManager")
BiocManager::install("DelayedMatrixStats")
```
## Example
This example compares two ways of computing column sums of a
*DelayedMatrix* object:
1. `DelayedMatrix::colSums()`: The ‘block-processing strategy’,
implemented in the **DelayedArray** package. The block-processing
strategy works for any *DelayedMatrix* object, regardless of the
type of *seed*.
2. `DelayedMatrixStats::colSums2()`: The ‘seed-aware’ strategy,
implemented in the **DelayedMatrixStats** package. The seed-aware
implementation is optimized for both speed and memory but only for
*DelayedMatrix* objects with certain types of *seed*.
<!-- end list -->
#> Warning: package 'sparseMatrixStats' was built under R version 4.0.3
``` r
library(DelayedMatrixStats)
library(sparseMatrixStats)
library(microbenchmark)
library(profmem)
```
``` r
set.seed(666)
# Fast column sums of DelayedMatrix with matrix seed
dense_matrix <- DelayedArray(matrix(runif(20000 * 600), nrow = 20000,
ncol = 600))
class(seed(dense_matrix))
#> [1] "matrix" "array"
dense_matrix
#> <20000 x 600> matrix of class DelayedMatrix and type "double":
#> [,1] [,2] [,3] ... [,599] [,600]
#> [1,] 0.7743685 0.6601787 0.4098798 . 0.89118118 0.05776471
#> [2,] 0.1972242 0.8436035 0.9198450 . 0.31799523 0.63099417
#> [3,] 0.9780138 0.2017589 0.4696158 . 0.31783791 0.02830454
#> [4,] 0.2013274 0.8797239 0.6474768 . 0.55217184 0.09678816
#> [5,] 0.3612444 0.8158778 0.5928599 . 0.08530977 0.39224147
#> ... . . . . . .
#> [19996,] 0.19490291 0.07763570 0.56391725 . 0.09703424 0.62659353
#> [19997,] 0.61182993 0.01910121 0.04046034 . 0.59708388 0.88389731
#> [19998,] 0.12932744 0.21155070 0.19344085 . 0.51682032 0.13378223
#> [19999,] 0.18985573 0.41716539 0.35110782 . 0.62939661 0.94601427
#> [20000,] 0.87889047 0.25308041 0.54666920 . 0.81630322 0.73272217
microbenchmark(DelayedArray::colSums(dense_matrix),
DelayedMatrixStats::colSums2(dense_matrix),
times = 10)
#> Unit: milliseconds
#> expr min lq mean
#> DelayedArray::colSums(dense_matrix) 58.10028 69.00577 143.70237
#> DelayedMatrixStats::colSums2(dense_matrix) 13.60972 15.11668 19.23493
#> median uq max neval cld
#> 81.08308 100.81742 414.00737 10 b
#> 16.10312 25.42459 30.91184 10 a
profmem::total(profmem::profmem(DelayedArray::colSums(dense_matrix)))
#> [1] 96105416
profmem::total(profmem::profmem(DelayedMatrixStats::colSums2(dense_matrix)))
#> [1] 166120
# Fast, low-memory column sums of DelayedMatrix with sparse matrix seed
sparse_matrix <- seed(dense_matrix)
zero_idx <- sample(length(sparse_matrix), 0.6 * length(sparse_matrix))
sparse_matrix[zero_idx] <- 0
sparse_matrix <- DelayedArray(Matrix::Matrix(sparse_matrix, sparse = TRUE))
class(seed(sparse_matrix))
#> [1] "dgCMatrix"
#> attr(,"package")
#> [1] "Matrix"
sparse_matrix
#> <20000 x 600> sparse matrix of class DelayedMatrix and type "double":
#> [,1] [,2] [,3] ... [,599] [,600]
#> [1,] 0.7743685 0.0000000 0.0000000 . 0.89118118 0.00000000
#> [2,] 0.1972242 0.0000000 0.9198450 . 0.00000000 0.00000000
#> [3,] 0.9780138 0.0000000 0.4696158 . 0.31783791 0.00000000
#> [4,] 0.0000000 0.8797239 0.6474768 . 0.55217184 0.00000000
#> [5,] 0.3612444 0.0000000 0.0000000 . 0.08530977 0.39224147
#> ... . . . . . .
#> [19996,] 0.1949029 0.0776357 0.0000000 . 0.09703424 0.00000000
#> [19997,] 0.0000000 0.0000000 0.0000000 . 0.00000000 0.88389731
#> [19998,] 0.0000000 0.2115507 0.1934408 . 0.00000000 0.00000000
#> [19999,] 0.1898557 0.0000000 0.3511078 . 0.62939661 0.94601427
#> [20000,] 0.8788905 0.2530804 0.0000000 . 0.00000000 0.73272217
microbenchmark(DelayedArray::colSums(sparse_matrix),
DelayedMatrixStats::colSums2(sparse_matrix),
times = 10)
#> Unit: milliseconds
#> expr min lq mean
#> DelayedArray::colSums(sparse_matrix) 250.41540 297.64721 504.14460
#> DelayedMatrixStats::colSums2(sparse_matrix) 15.45005 15.52633 17.30676
#> median uq max neval cld
#> 489.64900 639.84463 945.35155 10 b
#> 16.13531 16.41159 26.49614 10 a
profmem::total(profmem::profmem(DelayedArray::colSums(sparse_matrix)))
#> [1] 249647176
profmem::total(profmem::profmem(DelayedMatrixStats::colSums2(sparse_matrix)))
#> [1] 7400
# Fast column sums of DelayedMatrix with Rle-based seed
rle_matrix <- RleArray(Rle(sample(2L, 200000 * 6 / 10, replace = TRUE), 100),
dim = c(2000000, 6))
class(seed(rle_matrix))
#> [1] "SolidRleArraySeed"
#> attr(,"package")
#> [1] "DelayedArray"
rle_matrix
#> <2000000 x 6> matrix of class RleMatrix and type "integer":
#> [,1] [,2] [,3] [,4] [,5] [,6]
#> [1,] 2 2 1 1 1 2
#> [2,] 2 2 1 1 1 2
#> [3,] 2 2 1 1 1 2
#> [4,] 2 2 1 1 1 2
#> [5,] 2 2 1 1 1 2
#> ... . . . . . .
#> [1999996,] 1 2 2 1 1 1
#> [1999997,] 1 2 2 1 1 1
#> [1999998,] 1 2 2 1 1 1
#> [1999999,] 1 2 2 1 1 1
#> [2000000,] 1 2 2 1 1 1
microbenchmark(DelayedArray::colSums(rle_matrix),
DelayedMatrixStats::colSums2(rle_matrix),
times = 10)
#> Unit: milliseconds
#> expr min lq mean
#> DelayedArray::colSums(rle_matrix) 636.779156 659.794020 700.62470
#> DelayedMatrixStats::colSums2(rle_matrix) 4.482569 5.028546 14.81272
#> median uq max neval cld
#> 677.413206 699.000029 842.93055 10 b
#> 5.689509 6.471868 84.09537 10 a
profmem::total(profmem::profmem(DelayedArray::colSums(rle_matrix)))
#> [1] 168002536
profmem::total(profmem::profmem(DelayedMatrixStats::colSums2(rle_matrix)))
#> [1] 1640
```
## Benchmarking
An extensive set of benchmarks is under development at
<http://peterhickey.org/BenchmarkingDelayedMatrixStats/>.
## API coverage
- ✔ = Implemented in **DelayedMatrixStats**
- ☑️ = Implemented in
[**DelayedArray**](http://bioconductor.org/packages/DelayedArray/)
or
[**sparseMatrixStats**](http://bioconductor.org/packages/sparseMatrixStats/)
- ❌: = Not yet implemented
| Method | Block processing | *base::matrix* optimized | *Matrix::dgCMatrix* optimized | *Matrix::lgCMatrix* optimized | *DelayedArray::RleArray* (*SolidRleArraySeed*) optimized | *DelayedArray::RleArray* (*ChunkedRleArraySeed*) optimized | *HDF5Array::HDF5Matrix* optimized | *base::data.frame* optimized | *S4Vectors::DataFrame* optimized |
| :--------------------- | :--------------- | :----------------------- | :---------------------------- | :---------------------------- | :------------------------------------------------------- | :--------------------------------------------------------- | :-------------------------------- | :--------------------------- | :------------------------------- |
| `colAlls()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colAnyMissings()` | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colAnyNAs()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colAnys()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colAvgsPerRowSet()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colCollapse()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colCounts()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colCummaxs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colCummins()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colCumprods()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colCumsums()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colIQRDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colIQRs()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colLogSumExps()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colMadDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colMads()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colMaxs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colMeans2()` | ✔ | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ |
| `colMedians()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colMins()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colOrderStats()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colProds()` | ✔ | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ |
| `colQuantiles()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colRanges()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colRanks()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colSdDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colSds()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colsum()` | ☑️ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colSums2()` | ✔ | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ |
| `colTabulates()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colVarDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colVars()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colWeightedMads()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colWeightedMeans()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colWeightedMedians()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colWeightedSds()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `colWeightedVars()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowAlls()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowAnyMissings()` | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowAnyNAs()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowAnys()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowAvgsPerColSet()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowCollapse()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowCounts()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowCummaxs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowCummins()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowCumprods()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowCumsums()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowIQRDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowIQRs()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowLogSumExps()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowMadDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowMads()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowMaxs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowMeans2()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowMedians()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowMins()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowOrderStats()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowProds()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowQuantiles()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowRanges()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowRanks()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowSdDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowSds()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowsum()` | ☑️ | ❌ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowSums2()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowTabulates()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowVarDiffs()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowVars()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowWeightedMads()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowWeightedMeans()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowWeightedMedians()` | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowWeightedSds()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `rowWeightedVars()` | ✔ | ✔ | ✔ | ✔ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
