Halve the median runtime of gs_power_ahr()

[jdblischak]

Inspired by Issue #219, I was able to halve the runtime of the gs_power_ahr() example in the documentation for gs_spending_bound(). I achieved this by:

• converting ahr() to data.table
• converting almost all of pw_info() to data.table (there is a tricky group arrange operation at the end that I'd need to investigate more)
• updating expected_event() to return a data frame instead of a tibble
• updating pw_info() to pre-allocate lists instead of growing the data frame via rbind() within each loop iteration

While optimizing, I performed the following cleanups:

• added library(gsDesign) to tests/testthat/test-independent-gs_power_ahr.R so that it can be run via devtools::test()
• had to do some NAMESPACE rearranging to be able to use := from both the data.table and rlang packages while still keeping R CMD check happy
• added some missing @importFrom dplyr filter (I didn't get to all of them)

library("microbenchmark")

remotes::install_github("Merck/gsDesign2@50067c4", INSTALL_opts = "--with-keep.source")
library("gsDesign2")
packageVersion("gsDesign2")
## [1] ‘1.1.0’

microbenchmark(
  gs_power_ahr = gs_power_ahr(
    analysis_time = c(12, 24, 36),
    event = c(30, 40, 50),
    binding = TRUE,
    upper = gs_spending_bound,
    upar = list(sf = gsDesign::sfLDOF, total_spend = 0.025, param = NULL, timing = NULL),
    lower = gs_spending_bound,
    lpar = list(sf = gsDesign::sfLDOF, total_spend = 0.025, param = NULL, timing = NULL)
  ),
  times = 10
)
## Unit: milliseconds
##          expr      min       lq     mean   median       uq      max neval
##  gs_power_ahr 808.6144 840.0866 940.8113 925.5892 988.4552 1201.223    10

detach("package:gsDesign2")
devtools::load_all()
packageVersion("gsDesign2")
## [1] ‘1.1.0.1’

microbenchmark(
  gs_power_ahr = gs_power_ahr(
    analysis_time = c(12, 24, 36),
    event = c(30, 40, 50),
    binding = TRUE,
    upper = gs_spending_bound,
    upar = list(sf = gsDesign::sfLDOF, total_spend = 0.025, param = NULL, timing = NULL),
    lower = gs_spending_bound,
    lpar = list(sf = gsDesign::sfLDOF, total_spend = 0.025, param = NULL, timing = NULL)
  ),
  times = 10
)
## Unit: milliseconds
##          expr      min       lq     mean   median       uq      max neval
##  gs_power_ahr 405.3252 427.6431 507.1905 431.4219 538.8579 779.9118    10
925 / 431
## [1] 2.146172

[nanxstats]

This looks good from the technical perspective. We should discuss this in the meeting this week especially about using data.table.

[nanxstats]

Nice speed improvements. It's exciting that we have decided to use data.table to accelerate the gsDesign2 backend, too!

It seems the only explicit rlang usage (and := usage) is in R/utility_tidy_tbl.R:

dplyr::rename(!!aname := !!"_alab")

I think it would be ideal to use an alternative approach for the line above, so that 1. := is only used by data.table (ever) 2. rlang can be remove from Imports. 3. the ignore_unused_imports() workaround can be removed.