s3alfisc · s3alfisc · Sep 13, 2023 · Sep 13, 2023 · Sep 13, 2023 · Sep 13, 2023
diff --git a/CRAN-SUBMISSION b/CRAN-SUBMISSION
@@ -1,4 +1,3 @@
-Version: 0.14.0
-Date: 2023-07-09 19:55:48 UTC
-SHA:
-    431829b4d99cd13f0652e71483b3ce8e87cb5b58
+Version: 0.15.0
+Date: 2023-09-16 18:11:27 UTC
+SHA: 47b7f6ad8815300458eca67327699ea865a8c878
diff --git a/DESCRIPTION b/DESCRIPTION
@@ -1,6 +1,6 @@
 Package: fwildclusterboot
 Title: Fast Wild Cluster Bootstrap Inference for Linear Models
-Version: 0.14.3
+Version: 0.15.0
 Authors@R: c(
     person("Alexander", "Fischer", , "[email protected]", role = c("aut", "cre")),
     person("David", "Roodman", role = "aut"),

diff --git a/NAMESPACE b/NAMESPACE
@@ -30,11 +30,11 @@ export(glance)
 export(mboottest)
 export(pval)
 export(setBoottest_engine)
+export(setBoottest_nthreads)
 export(teststat)
 export(tidy)
 importFrom(JuliaConnectoR,juliaEval)
 importFrom(JuliaConnectoR,juliaImport)
-importFrom(Matrix,Diagonal)
 importFrom(Matrix,Matrix)
 importFrom(Matrix,crossprod)
 importFrom(Matrix,sparse.model.matrix)
@@ -64,6 +64,7 @@ importFrom(rlang,warn)
 importFrom(stats,aggregate)
 importFrom(stats,as.formula)
 importFrom(stats,coef)
+importFrom(stats,confint)
 importFrom(stats,expand.model.frame)
 importFrom(stats,formula)
 importFrom(stats,model.frame)

diff --git a/NEWS.md b/NEWS.md
@@ -1,3 +1,39 @@
+# CRAN release: fwildclusterboot 0.15
+
+- For rademacher weights, switches to `dqrng::dqrrademacher`. This leads to substabtial 
+  performance increases (see below). But the random number generation changes slightly, so 
+  you might no longer reproduce exactly the same weights draw under the same seeds, and in turn, 
+  your bootstrap inferences might change slightly. 
+- For other changes, see the changelog for versions `0.13-0.14.3`, all of which have not 
+  been published on CRAN. 
+- The package's lifecycle badge has been changed to `stable`, as there have not been any breaking API 
+  changes in a while, and I do not plan any in the future. 
+
+```r
+  > library(bench)
+> 
+> n <- 999999 * 100
+> 
+> mark(
++   dqrrademacher(n), 
++   dqrng::dqsample(
++     x = c(-1L, 1L),
++     size = n,
++     replace = TRUE
++   ), 
++   iterations = 5, 
++   check = FALSE
++ )
+# A tibble: 2 × 13
+  expression                                                   min median itr/s…¹ mem_a…² gc/se…³ n_itr  n_gc
+  <bch:expr>                                                <bch:> <bch:>   <dbl> <bch:b>   <dbl> <int> <dbl>
+1 dqrrademacher(n)                                           123ms  166ms    5.25   381MB    3.15     5     3
+2 dqrng::dqsample(x = c(-1L, 1L), size = n, replace = TRUE)  615ms  647ms    1.47   763MB    1.47     5     5
+# … with 5 more variables: total_time <bch:tm>, result <list>, memory <list>, time <list>, gc <list>, and
+#   abbreviated variable names ¹`itr/sec`, ²mem_alloc, ³`gc/sec`
+```
+
+
 # fwildclusterboot 0.14.3
 
 - Fix a bug with CI inversion when `r` was set to be close to the estimated parameter. (CI inversion failed). See [#138](https://github.com/s3alfisc/fwildclusterboot/issues/138). Thanks to Achim Zeileis & team for raising this issue!

diff --git a/R/boot_algo_fastnwild.R b/R/boot_algo_fastnwild.R
@@ -312,9 +312,9 @@ boot_algo_fastnwild <-
           if (impose_null == TRUE) {
             # should always be positive, point_estimate and t_stat need to have same
             # sign, abs for security
-            pval_peak <- abs(point_estimate)
+            pval_peak <- point_estimate
           } else if (impose_null == FALSE) {
-            pval_peak <- abs((point_estimate - r))
+            pval_peak <- (point_estimate - r)
           }
 
           invert_p_val(

diff --git a/R/boottest_felm.R b/R/boottest_felm.R
@@ -421,6 +421,12 @@ boottest.felm <- function(object,
       }
     }
   }
+
+  check_engine_btype(
+    engine = engine,
+    bootstrap_type = bootstrap_type
+  )
+
 
   if (!is.null(beta0)) {
     rlang::abort(
@@ -454,14 +460,18 @@ boottest.felm <- function(object,
 
   if (is.null(clustid)) {
     heteroskedastic <- TRUE
-    if (engine == "R") {
+    if (engine %in% c("R", "WildBootTests.jl")) {
       # heteroskedastic models should always be run through R-lean
       engine <- "R-lean"
+      rlang::inform(
+        "The heteroskedastic bootstrap is only supported for `engine = 'R-lean'`, 
+        Resetting the engine argument to 'R-lean'."
+      )
     }
   } else {
     heteroskedastic <- FALSE
   }
-
+  
   check_bootstrap_types(
     param = param,
     bootstrap_type = bootstrap_type
@@ -492,7 +502,7 @@ boottest.felm <- function(object,
     fe = fe
   )
 
-  preprocess <- preprocess2.felm(
+  preprocess <- preprocess_felm(
     object = object,
     clustid = clustid,
     R = R_long,

diff --git a/R/boottest_fixest.R b/R/boottest_fixest.R
@@ -431,7 +431,12 @@ boottest.fixest <- function(object,
       }
     }
   }
-
+
+  check_engine_btype(
+    engine = engine,
+    bootstrap_type = bootstrap_type
+  )
+
   if (!is.null(beta0)) {
     rlang::abort(
       c(
@@ -481,14 +486,18 @@ boottest.fixest <- function(object,
 
   if (is.null(clustid)) {
     heteroskedastic <- TRUE
-    if (engine == "R") {
+    if (engine %in% c("R", "WildBootTests.jl")) {
       # heteroskedastic models should always be run through R-lean
       engine <- "R-lean"
+      rlang::inform(
+        "The heteroskedastic bootstrap is only supported for `engine = 'R-lean'`, 
+        Resetting the engine argument to 'R-lean'."
+      )
     }
   } else {
     heteroskedastic <- FALSE
   }
-
+  
   check_bootstrap_types(
     param = param,
     bootstrap_type = bootstrap_type
@@ -521,7 +530,7 @@ boottest.fixest <- function(object,
   )
 
   # preprocess the data: Y, X, weights, fixed_effect
-  preprocess <- preprocess2.fixest(
+  preprocess <- preprocess_fixest(
     object = object,
     clustid = clustid,
     R = R_long,

diff --git a/R/boottest_ivreg.R b/R/boottest_ivreg.R
@@ -288,7 +288,7 @@ boottest.ivreg <- function(object,
 
 
   # preprocess data: X, Y, weights, fixed effects
-  preprocess <- preprocess2.ivreg(
+  preprocess <- preprocess_ivreg(
     object = object,
     clustid = clustid,
     R = R,

diff --git a/R/boottest_lm.R b/R/boottest_lm.R
@@ -383,7 +383,7 @@ boottest.lm <- function(object,
     engine = engine,
     bootstrap_type = bootstrap_type
   )
-
+  
   if (!is.null(beta0)) {
     rlang::abort(
       c(
@@ -413,14 +413,18 @@ boottest.lm <- function(object,
 
   if (is.null(clustid)) {
     heteroskedastic <- TRUE
-    if (engine == "R") {
+    if (engine %in% c("R", "WildBootTests.jl")) {
       # heteroskedastic models should always be run through R-lean
       engine <- "R-lean"
+      rlang::inform(
+        "The heteroskedastic bootstrap is only supported for `engine = 'R-lean'`, 
+        Resetting the engine argument to 'R-lean'."
+      )
     }
   } else {
     heteroskedastic <- FALSE
   }
-
+    
   check_bootstrap_types(
     param = param,
     bootstrap_type = bootstrap_type
@@ -453,7 +457,7 @@ boottest.lm <- function(object,
   # now split into R, R-lean and WildBootTests.jl algos
   # different pre-processing and different algo-functions
 
-  preprocess <- preprocess2.lm(
+  preprocess <- preprocess_lm(
     object = object,
     clustid = clustid,
     R = R_long,

diff --git a/R/get_weights.R b/R/get_weights.R
@@ -29,11 +29,7 @@ get_weights <- function(type,
                             # (uses less memory
                             # than numeric)
                             rademacher = function(n) {
-                              dqrng::dqsample(
-                                x = c(-1L, 1L),
-                                size = n,
-                                replace = TRUE
-                              )
+                              dqrng::dqrrademacher(n)
                             },
                             mammen = function(n) {
                               sample(

diff --git a/R/mboottest_felm.R b/R/mboottest_felm.R
@@ -229,7 +229,7 @@ mboottest.felm <- function(object,
 
   fedfadj <- 0L
 
-  preprocess <- preprocess2.felm(
+  preprocess <- preprocess_felm(
     object = object,
     clustid = clustid,
     R = R,

diff --git a/R/mboottest_fixest.R b/R/mboottest_fixest.R
@@ -255,7 +255,7 @@ mboottest.fixest <- function(object,
     fe = fe
   )
 
-  preprocess <- preprocess2.fixest(
+  preprocess <- preprocess_fixest(
     object = object,
     clustid = clustid,
     R = R,

diff --git a/R/mboottest_lm.R b/R/mboottest_lm.R
@@ -226,7 +226,7 @@ mboottest.lm <- function(object,
 
 
   # preprocess data: X, Y, weights, fixed effects
-  preprocess <- preprocess2.lm(
+  preprocess <- preprocess_lm(
     object = object,
     clustid = clustid,
     R = R,

diff --git a/R/methods.R b/R/methods.R
@@ -52,7 +52,7 @@
 #' 
 #' @details Technical Details 
 #' For technical details, either take a look at the references below, or check
-#' out the [wild (cluster) bootstrap vignette](wild_bootstrap.html).
+#' out the [wild (cluster) bootstrap vignette](https://s3alfisc.github.io/fwildclusterboot/articles/wild_bootstrap.html).
 #' 
 #' @section Stata, Julia and Python Implementations:
 #' The fast wild cluster bootstrap algorithms are further implemented in the
@@ -757,6 +757,7 @@ nobs.mboottest <- function(object, ...) {
 #' @param ... Further arguments passed to or from other methods.
 #' @param digits Number of rounding digits
 #' @export
+#' @importFrom stats confint
 #' @method print boottest
 #' @return A scalar containing the effective number of observations
 #' used in `mboottest`
@@ -780,16 +781,17 @@ print.boottest <- function(x, ..., digits = 4) {
   print(x$call)
   cat("", "\n")
 
-  vals <- lapply(
-    c("p_val", "conf_int", "t_stat"),
-    function(y) {
-      round(x[[y]], digits = digits)
-    }
-  )
+  pval <- pval(x)
+  conf_int <- confint(x)
+  teststat <- teststat(x)
 
-  cat("p value:", vals[[1]], "\n")
-  cat("confidence interval:", vals[[2]], "\n")
-  cat("test statistic", vals[[3]], "\n")
+  pval <- round(pval, digits = digits)
+  teststat <- round(teststat, digits = digits)
+  conf_int <- ifelse(is.null(conf_int), NA, round(conf_int, digits))
+
+  cat("p value:", pval, "\n")
+  cat("confidence interval:", conf_int, "\n")
+  cat("test statistic", teststat, "\n")
 }
 
 

diff --git a/R/onLoad.R b/R/onLoad.R
@@ -16,8 +16,17 @@
 
 .onAttach <- 
 function(libname, pkgname) {
+
+  packageStartupMessage("\nNote that with version 0.15, rademacher weights are")
+  packageStartupMessage("drawn from `dqrng::dqrrademacher()` instead of `dqrng::dqsample()`.")
+  packageStartupMessage("This brings nice performance gains, but you might no longer be able to exactly reproduce")
+  packageStartupMessage("your bootstrap inferences under a given seed. If this is a ")
+  packageStartupMessage("requirement for you, please revert to `fwildclusterboot` version")
+  packageStartupMessage("0.13-0.14.3.")
+
   packageStartupMessage("\nPlease cite as: \n")
   packageStartupMessage(" Fischer & Roodman. (2021). fwildclusterboot: Fast Wild Cluster.")
   packageStartupMessage(" Bootstrap Inference for Linear Regression Models.")
   packageStartupMessage(" Available from https://cran.r-project.org/package=fwildclusterboot/.")
+
 }
diff --git a/R/preprocess2.R → R/preprocess_models.R b/R/preprocess2.R → R/preprocess_models.R
@@ -1,20 +1,4 @@
-#' `preprocess2` is a S3 method that fetches data from several model
-#' objectects for use with `boottest()`.
-#'
-#' @param object An objectect of type lm, fixest, felm or ivreg
-#' @param ... other arguments
-#' @noRd
-#'
-#' @importFrom Matrix Diagonal
-#' @return An object of class `preprocess2`.
-#' @srrstats {G2.8} *Software should provide appropriate conversion or dispatch
-#'  routines as part of initial pre-processing to ensure that all other
-#'  sub-functions of a package receive inputs of a single defined class or
-#'  type.* Similar preprocess method + run_bootstrap function.
-preprocess2 <- function(object, ...) {
-  UseMethod("preprocess2")
-}
-preprocess2.fixest <-
+preprocess_fixest <-
   function(object,
            clustid,
            R,
@@ -213,7 +197,7 @@ preprocess2.fixest <-
     }
     res
   }
-preprocess2.felm <-
+preprocess_felm <-
   function(object,
            clustid,
            R,
@@ -388,7 +372,7 @@ preprocess2.felm <-
     }
     res
   }
-preprocess2.lm <-
+preprocess_lm <-
   function(object,
            clustid,
            R,
@@ -492,7 +476,7 @@ preprocess2.lm <-
     class(res) <- c("preprocess", "ols")
     res
   }
-preprocess2.ivreg <-
+preprocess_ivreg <-
   function(object,
            clustid,
            R,
@@ -1040,4 +1024,4 @@ get_cluster <-
       cluster_bootcluster = cluster_bootcluster_df
     )
     res
-}
+}