document how to run in parallel

R/multi.R

@@ -370,7 +370,7 @@ qmulti_est <- function(x, dist, p) {
 
 qmulti_ranges <- function(p, list) {
   elist <- purrr::imap(list, qmulti_est, p = p)
-  tlist <- purrr::transpose(elist)
+  tlist <- purrr::list_transpose(elist)
   tlist <- purrr::map(tlist, unlist)
   min <- purrr::map_dbl(tlist, min)
   max <- purrr::map_dbl(tlist, max)

R/test-helpers.R

@@ -12,11 +12,11 @@
 #    See the License for the specific language governing permissions and
 #    limitations under the License.
 
-local_multisession <- function(.local_envir = parent.frame()) {
+local_multisession <- function(.local_envir = parent.frame(), workers = 2) {
   oldDoPar <- doFuture::registerDoFuture()
-  withr::defer(with(oldDoPar, foreach::setDoPar(fun = fun, data = data, info = info)))
-  oldPlan <- future::plan("future::multisession")
-  withr::defer(future::plan(oldPlan))
+  withr::defer_parent(with(oldDoPar, foreach::setDoPar(fun = fun, data = data, info = info)))
+  oldPlan <- future::plan("future::multisession", workers = workers)
+  withr::defer_parent(future::plan(oldPlan))
   invisible(oldDoPar)
 }
 

README.Rmd

@@ -52,7 +52,7 @@ ssddata::ccme_boron
 
 Distributions are fit using `ssd_fit_dists()`
 ```{r}
-fits <- ssd_fit_dists(ssddata::ccme_boron)
+fits <- ssd_fit_dists(ssddata::ccme_boron, dists = c("lnorm", "llogis"))
 ```
 
 and can be quickly plotted using `autoplot`
@@ -70,13 +70,19 @@ The goodness of fit can be assessed using `ssd_gof`
 ssd_gof(fits)
 ```
 
-and the model-averaged 5% hazard concentration estimated by parametric bootstrapping using `ssd_hc`
+and the model-averaged 5% hazard concentration estimated by parametric bootstrapping using `ssd_hc`.
 ```{r}
 set.seed(99)
 hc5 <- ssd_hc(fits, ci = TRUE, nboot = 100) # 100 bootstrap samples for speed
 print(hc5)
 ```
 
+To bootstrap in parallel set `future::plan()`. For example:
+```{r, eval=FALSE}
+future::multisession(workers = 2)
+hc5 <- ssd_hc(fits, ci = TRUE, nboot = 100)
+```
+
 Model-averaged predictions complete with confidence intervals can also be estimated by parametric bootstrapping using the `stats` generic `predict`.
 To perform bootstrapping for each distribution in parallel register the future backend
 and then select the evaluation strategy.

README.md

@@ -68,7 +68,7 @@ ssddata::ccme_boron
 Distributions are fit using `ssd_fit_dists()`
 
 ``` r
-fits <- ssd_fit_dists(ssddata::ccme_boron)
+fits <- ssd_fit_dists(ssddata::ccme_boron, dists = c("lnorm", "llogis"))
 ```
 
 and can be quickly plotted using `autoplot`
@@ -88,19 +88,15 @@ The goodness of fit can be assessed using `ssd_gof`
 
 ``` r
 ssd_gof(fits)
-#> # A tibble: 6 × 9
-#>   dist           ad     ks    cvm   aic  aicc   bic delta weight
-#>   <chr>       <dbl>  <dbl>  <dbl> <dbl> <dbl> <dbl> <dbl>  <dbl>
-#> 1 gamma       0.440 0.117  0.0554  238.  238.  240. 0.005  0.357
-#> 2 lgumbel     0.829 0.158  0.134   244.  245.  247. 6.56   0.013
-#> 3 llogis      0.487 0.0994 0.0595  241.  241.  244. 3.39   0.066
-#> 4 lnorm       0.507 0.107  0.0703  239.  240.  242. 1.40   0.177
-#> 5 lnorm_lnorm 0.320 0.116  0.0414  240.  243.  247. 4.98   0.03 
-#> 6 weibull     0.434 0.117  0.0542  238.  238.  240. 0      0.357
+#> # A tibble: 2 × 9
+#>   dist      ad     ks    cvm   aic  aicc   bic delta weight
+#>   <chr>  <dbl>  <dbl>  <dbl> <dbl> <dbl> <dbl> <dbl>  <dbl>
+#> 1 lnorm  0.507 0.107  0.0703  239.  240.  242.  0      0.73
+#> 2 llogis 0.487 0.0994 0.0595  241.  241.  244.  1.99   0.27
 ```
 
 and the model-averaged 5% hazard concentration estimated by parametric
-bootstrapping using `ssd_hc`
+bootstrapping using `ssd_hc`.
 
 ``` r
 set.seed(99)
@@ -109,7 +105,14 @@ print(hc5)
 #> # A tibble: 1 × 10
 #>   dist    percent   est    se   lcl   ucl    wt method     nboot pboot
 #>   <chr>     <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <chr>      <dbl> <dbl>
-#> 1 average       5  1.24 0.741 0.510  3.35     1 parametric   100     1
+#> 1 average       5  1.65 0.599 0.923  3.34     1 parametric   100     1
+```
+
+To bootstrap in parallel set `future::plan()`. For example:
+
+``` r
+future::multisession(workers = 2)
+hc5 <- ssd_hc(fits, ci = TRUE, nboot = 100)
 ```
 
 Model-averaged predictions complete with confidence intervals can also
@@ -136,7 +139,7 @@ ssd_plot(ssddata::ccme_boron, boron_pred,
   scale_colour_ssd()
 ```
 
-![](man/figures/README-unnamed-chunk-9-1.png)<!-- -->
+![](man/figures/README-unnamed-chunk-10-1.png)<!-- -->
 
 ## References
 

tests/testthat/_snaps/hc/hcici.csv

@@ -0,0 +1,2 @@
+dist,percent,est,se,lcl,ucl,wt,method,nboot,pboot
+average,5,1.68117,0.455819,0.998397,2.43078,1,parametric,10,1

tests/testthat/_snaps/hc/hcici_multi.csv

@@ -0,0 +1,2 @@
+dist,percent,est,se,lcl,ucl,wt,method,nboot,pboot
+average,5,1.68117,0.455819,0.998397,2.43078,1,parametric,10,1

tests/testthat/test-hc.R

@@ -321,7 +321,7 @@ test_that("ssd_hc calculates cis with equally weighted data", {
   fits <- ssd_fit_dists(data, weight = "Weight", dists = "lnorm")
   set.seed(10)
   hc <- ssd_hc(fits, ci = TRUE, nboot = 10, multi = FALSE)
-  expect_equal(hc$se, 0.455819097122445)
+  expect_snapshot_data(hc, "hcici")
 })
 
 test_that("ssd_hc calculates cis in parallel but one distribution", {
@@ -330,7 +330,7 @@ test_that("ssd_hc calculates cis in parallel but one distribution", {
   fits <- ssd_fit_dists(data, dists = "lnorm")
   set.seed(10)
   hc <- ssd_hc(fits, ci = TRUE, nboot = 10, multi = FALSE)
-  expect_equal(hc$se, 0.455819097122445)
+  expect_snapshot_data(hc, "hcici_multi")
 })
 
 test_that("ssd_hc calculates cis with two distributions", {
@@ -675,3 +675,14 @@ test_that("not all estimates if fail", {
                                       "estimates_000000005_multi.rds", "estimates_000000006_multi.rds", 
                                       "estimates_000000007_multi.rds", "estimates_000000009_multi.rds", "estimates_000000010_multi.rds"))
 })
+
+test_that("ssd_hc identical if in parallel", {
+  data <- ssddata::ccme_boron
+  fits <- ssd_fit_dists(data, dists = c("lnorm", "llogis"))
+  set.seed(10)
+  hc <- ssd_hc(fits, ci = TRUE, nboot = 500)
+  local_multisession(workers = 2)
+  set.seed(10)
+  hc2 <- ssd_hc(fits, ci = TRUE, nboot = 500)
+  expect_identical(hc, hc2)
+})