Benchmarks for M/Z sorted Spectrum1

[jorainer]

Wanted to share benchmarks for different approaches to ensure/enforce ordering of M/Z-intensity value pairs by M/Z value in Spectrum1 objects (relates to issue #135):

I compared:

• Sorting of values in R (after reading the values with mzR and before creating the Spectrum1 objects) using either method = "radix" and method = "shell".
• Sorting of values in the C constructor for Spectrum1 objects (i.e. sorting values in C).

> suppressWarnings(
+     onDisk <- readMSData2(files = mzf, centroided = TRUE)
+ )
Reading 3670 spectra from file POS_C_O_001.mzML
Reading 3670 spectra from file POS_C_O_002.mzML
> library(microbenchmark)
> fDataPerFile <- split(fData(onDisk), f = fData(onDisk)$fileIdx)
> BPPARAM <- SerialParam()
> ## Without M/Z sorting
> unsorted <- function() {
+         bplapply(fDataPerFile,
+                  FUN = MSnbase:::.applyFun2SpectraOfFileMulti,
+                  filenames = fileNames(onDisk), queue = list(),
+                  APPLYFUN = NULL, BPPARAM = BPPARAM)
+ }
> ## With sorting in R using method = "radix"
> radixSortedR <- function() {
+     bplapply(fDataPerFile,
+              FUN = MSnbase:::.applyFun2SpectraOfFileMultiRadixSortMz,
+              filenames = fileNames(onDisk), queue = list(),
+              APPLYFUN = NULL, BPPARAM = BPPARAM)
+ }
> ## With sorting in R using method = "shell"
> shellSortedR <- function() {
+     bplapply(fDataPerFile,
+              FUN = MSnbase:::.applyFun2SpectraOfFileMultiShellSortMz,
             filenames = fileNames(onDisk), queue = list(),
+ +              APPLYFUN = NULL, BPPARAM = BPPARAM)
+ }
> ## With sorting performed in the C constructor
> constructorSorted <- function() {
+     bplapply(fDataPerFile,
+              FUN = MSnbase:::.applyFun2SpectraOfFileMulti2,
+              filenames = fileNames(onDisk), queue = list(),
+              APPLYFUN = NULL, BPPARAM = BPPARAM)
+ }
>
> ## o Compare unsorted vs radix sorted in R
> microbenchmark(unsorted(), radixSortedR(),times = 10)
Unit: seconds
           expr     min       lq     mean   median        uq       max neval
     unsorted() 5.82697 6.225438 6.769096 6.859658  7.058579  8.187952    10
 radixSortedR() 7.45798 7.800574 8.950423 8.530921 10.016772 11.586625    10
 cld
  a 
   b
> 
> ## o Compare radix sorted in R vs shell sorted in R
> microbenchmark(radixSortedR(), shellSortedR(), times = 10)
Unit: seconds
           expr      min       lq     mean   median       uq       max neval
 radixSortedR() 7.962823 8.044155 8.680846 8.381478 8.951652 10.734356    10
 shellSortedR() 8.145214 8.515322 8.846406 8.711967 9.217153  9.994807    10
 cld
   a
   a
> 
> ## o Compare radix sorted in R vs sorted in C constructor
> microbenchmark(radixSortedR(), constructorSorted(), times = 10)
Unit: seconds
                expr      min       lq     mean   median       uq      max
      radixSortedR() 7.401449 7.435909 7.847040 7.779204 8.147483 8.541478
 constructorSorted() 6.189033 6.503043 6.616889 6.641096 6.725856 7.055754
 neval cld
    10   b
    10  a 

There's no big difference, but still, the implementation in C outperforms all.