Benchmarking improvements in rhdf5 index to hyperslab selection

benchmark_index2hyperslab.R

BiocManager::install("grimbough/rhdf5", ref = "f06ab6f",
                     update = FALSE, ask = TRUE, INSTALL_opts = c('--no-lock'))

suppressPackageStartupMessages(library(rhdf5))
suppressPackageStartupMessages(library(microbenchmark))
suppressPackageStartupMessages(library(ggplot2))
suppressPackageStartupMessages(library(dplyr))
suppressPackageStartupMessages(library(stringr))

h5file <- "/tmpdata/msmith/ExperimentHub/2230"
dm <- h5read(h5file, name = "assay001")

f1 <- function(i, type) {
    
    row_index <- switch(type, 
                    seq = seq(1,nrow(dm), length.out = i),
                    rand = sample(1:nrow(dm), i),
                    lin = 1:i)
    
    h5read(h5file, name = "assay001", index = list(row_index, NULL))
}

bm_old <- microbenchmark(
    f1(100, 'seq'),  f1(500, 'seq'),  f1(1000, 'seq'),  f1(2000, 'seq'),  f1(3000, 'seq'),  f1(4500, 'seq'), 
    f1(6000, 'seq'), f1(7000, 'seq'), f1(8000, 'seq'), f1(9000, 'seq'), f1(10000, 'seq'), f1(12000, 'seq'),
    f1(100, 'rand'),  f1(500, 'rand'),  f1(1000, 'rand'),  f1(2000, 'rand'), f1(3000, 'rand'), f1(4500, 'rand'), 
    f1(6000, 'rand'), f1(7000, 'rand'), f1(8000, 'rand'), f1(9000, 'rand'), f1(10000, 'rand'), f1(12000, 'rand'),
    f1(100, 'lin'),  f1(500, 'lin'),  f1(1000, 'lin'),  f1(2000, 'lin'), f1(3000, 'lin'), f1(4500, 'lin'), 
    f1(6000, 'lin'), f1(7000, 'lin'), f1(8000, 'lin'), f1(9000, 'lin'), f1(10000, 'lin'), f1(12000, 'lin'),
    times = 8
)

res1 <- bm_old %>% mutate(index_length = str_extract(expr, "[0-9]{3,}") %>% as.integer(), 
                          type = str_extract(expr, "[a-z]{3,}"),
                          rhdf5_version = Biobase::package.version('rhdf5')) %>%
    select(-expr) %>%
    as_tibble()

message(Biobase::package.version('rhdf5'))

set.seed(1234)
rand_old <- f1(i = 10000, type = "rand")

############################

unloadNamespace('rhdf5')
BiocManager::install('grimbough/rhdf5', ref = "02fcc60", 
                     update = FALSE, ask = TRUE, INSTALL_opts = c('--no-lock'))
library(rhdf5)

message(Biobase::package.version('rhdf5'))

bm_new <- microbenchmark(
    f1(100, 'seq'),  f1(500, 'seq'),  f1(1000, 'seq'),  f1(2000, 'seq'),  f1(3000, 'seq'),  f1(4500, 'seq'), 
    f1(6000, 'seq'), f1(7000, 'seq'), f1(8000, 'seq'), f1(9000, 'seq'), f1(10000, 'seq'), f1(12000, 'seq'),
    f1(100, 'rand'),  f1(500, 'rand'),  f1(1000, 'rand'),  f1(2000, 'rand'), f1(3000, 'rand'), f1(4500, 'rand'), 
    f1(6000, 'rand'), f1(7000, 'rand'), f1(8000, 'rand'), f1(9000, 'rand'), f1(10000, 'rand'), f1(12000, 'rand'),
    f1(100, 'lin'),  f1(500, 'lin'),  f1(1000, 'lin'),  f1(2000, 'lin'), f1(3000, 'lin'), f1(4500, 'lin'), 
    f1(6000, 'lin'), f1(7000, 'lin'), f1(8000, 'lin'), f1(9000, 'lin'), f1(10000, 'lin'), f1(12000, 'lin'),
    times = 8
)

res2 <- bm_new %>% mutate(index_length = str_extract(expr, "[0-9]{3,}") %>% as.integer(), 
                          type = str_extract(expr, "[a-z]{3,}"),
                          rhdf5_version = Biobase::package.version('rhdf5')) %>%
    select(-expr) %>%
    as_tibble()

set.seed(1234)
rand_new <- f1(i = 10000, type = "rand")

## check we have the same objects from both versions
message(identical(rand_old, rand_new))

## plot time taken
ggplot(bind_rows(res1, res2), aes(x = index_length, y = time / 1e9)) + 
       geom_point(aes(color = type, shape = rhdf5_version)) + 
       geom_smooth(aes(linetype = rhdf5_version, color = type), se = FALSE) +
    theme_bw()

Nice power-ups to the benchmarking code!