Chapter 7 review (csgillespie#161)

* Addressed reviewers comments.

* Use devtools version of package

* Trying to fix travis error

* Tracking down travis

* Still git dramas

* Bump Rcpp version

* Clear Glob Env

DESCRIPTION

@@ -13,7 +13,7 @@ Depends: assertive.reflection, benchmarkme, bookdown, cranlogs, data.table (>= 1
   devtools (>= 1.12.0), DiagrammeR, dplyr, drat, efficient, formatR, fortunes, 
   geosphere, ggmap, ggplot2 (>= 2.1.0), ggplot2movies, knitr, lubridate, 
   microbenchmark, profvis (>= 0.3.2), pryr, readr, RSQLite, tibble (>= 1.1.0), tidyr, 
-  Rcpp (>= 0.12.6)
+  Rcpp (>= 0.12.7)
 LazyData: TRUE
 LinkingTo: Rcpp
 Remotes:

code/07-performance_f3.R

@@ -1,59 +1,41 @@
 source("code/initialise.R")
-
-# # Code to download logs of various packages
-# # Not run to avoid cranlogs dependency for book
-# dd = cranlogs::cran_downloads(packages = c("V8", "Rcpp", "rPython", "rJava"),
-#                               from = "2013-01-01", to = "2016-09-01")
-# dd$Downloads <- ave(
-#   dd$count,
-#   dd$package,
-#   FUN = function(x)
-#     zoo::rollmean(x, k = 30, na.pad = T)
-# )
-# saveRDS(dd, "extdata/cranlog.Rds")
-
-dd = readRDS("extdata/cranlog.Rds")
-dd = dd[dd$count > 0, ]
-
-v8 = dd[dd$package=="V8",]
-rcpp = dd[dd$package=="Rcpp",]
-rjava = dd[dd$package=="rJava",]
-rpython = dd[dd$package=="rPython",]
+# ## Original idea from https://github.com/csgillespie/efficientR/issues/121 and @adamryczkowski
+# 
+# n_rep=10 #Number of times to generate a test sample. This smooths the chart.
+# dd = matrix(0, nrow=100, ncol=3)
+# ns = 10^(seq(from=log10(10), to=log10(10^6), length.out =nrow(dd)))
+# 
+# tmp = matrix(0, nrow=n_rep, ncol=2)
+# for (i in seq_len(nrow(dd)))
+# {
+#   for(j in seq_len(n_rep))
+#   {
+#     s = sample(c(TRUE, FALSE), ns[i], TRUE, prob=c(1-0.0001,0.0001))
+#     tmp[j,] = summary(
+#      microbenchmark::microbenchmark(
+#         which.max(s),
+#         which(s)[[1]])
+#       )[['median']]
+#   }
+#   dd[i,] = c(ns[i], colMeans(tmp))
+#   cat(".")
+# }
+#saveRDS(dd, file="extdata/07-which_comparison.RData")
+dd = readRDS(file="extdata/07-which_comparison.RData")
 
 par(mar=c(3,3,2,1), mgp=c(2,0.4,0), tck=-.01,
     cex.axis=0.9, las=1, xaxs='i',yaxs='i')
-
-# Blank graph
-start = as.Date("2013-01-01"); end = as.Date("2016-09-01")
-plot(rcpp$date[1], type="n", xlim=c(start, end), ylim=c(10^0, 2*10^4), 
+plot(dd[,1], dd[,3]/dd[,2], log="xy",
+     pch=21, xlim=c(9, 10^4), ylim=c(1, 10^2), 
      axes=FALSE, frame=FALSE, 
-     xlab="Time", ylab="Downloads per day", 
-     log="y")
-abline(h=10^(0:4), lty=3, col="grey80")
-
-# Add R versions
-R_dates = as.Date(c("2013-04-03", "2014-04-03", "2015-04-03", "2016-04-03"))
-R_vers = paste0("R 3.", 0:3)
-abline(v=R_dates, col="grey90", lty=2)
-text(R_dates, 11800, R_vers,  pos=4, col="grey50")
+     ylab="Relative speed of 'which.min' compared to 'which'", xlab="Vector length",
+     bg="steelblue")
+abline(h=10^(0:2), lty=3, col="grey80")
+abline(h=c(2, 5, 20, 50), lty=3, col="grey90")
 
-# Label lines
-lines(rcpp$date, rcpp$Downloads,  lwd=3,col=3)
-lines(v8$date, v8$Downloads, lwd=2, col=4)
-lines(rjava$date, rjava$Downloads, lwd=2, col=2)
-lines(rpython$date, rpython$Downloads, lwd=2, col=1)
 
-# Add axis
 axis(2, tick=FALSE)
-axis(1, at = c(as.Date("2013-01-01"),as.Date("2014-01-01"),as.Date("2015-01-01"), as.Date("2016-01-01")), 
-     labels=2013:2016, tick=F)
+axis(1, at = 10^(1:4), labels=c(expression(10^1),expression(10^2),expression(10^3),expression(10^4)), tick=F)
 
-title("The rise of Rcpp", adj=1, 
-      cex.main=1, font.main=2, col.main="black")
 
-# Label lines
-text(as.Date("2013-06-01"), 6, "rPython", col=1, font = 2)
-text(as.Date("2014-02-01"), 3000, "Rcpp", col=3, font = 2)
-text(as.Date("2015-02-01"), 80, "V8", col=4, font = 2)
-text(as.Date("2015-03-01"), 1000, "rJava", col=2, font = 2)
 

code/07-performance_f5.R

@@ -0,0 +1,59 @@
+source("code/initialise.R")
+
+# # Code to download logs of various packages
+# # Not run to avoid cranlogs dependency for book
+# dd = cranlogs::cran_downloads(packages = c("V8", "Rcpp", "rPython", "rJava"),
+#                               from = "2013-01-01", to = "2016-09-01")
+# dd$Downloads <- ave(
+#   dd$count,
+#   dd$package,
+#   FUN = function(x)
+#     zoo::rollmean(x, k = 30, na.pad = T)
+# )
+# saveRDS(dd, "extdata/cranlog.Rds")
+
+dd = readRDS("extdata/cranlog.Rds")
+dd = dd[dd$count > 0, ]
+
+v8 = dd[dd$package=="V8",]
+rcpp = dd[dd$package=="Rcpp",]
+rjava = dd[dd$package=="rJava",]
+rpython = dd[dd$package=="rPython",]
+
+par(mar=c(3,3,2,1), mgp=c(2,0.4,0), tck=-.01,
+    cex.axis=0.9, las=1, xaxs='i',yaxs='i')
+
+# Blank graph
+start = as.Date("2013-01-01"); end = as.Date("2016-09-01")
+plot(rcpp$date[1], type="n", xlim=c(start, end), ylim=c(10^0, 2*10^4), 
+     axes=FALSE, frame=FALSE, 
+     xlab="Time", ylab="Downloads per day", 
+     log="y")
+abline(h=10^(0:4), lty=3, col="grey80")
+
+# Add R versions
+R_dates = as.Date(c("2013-04-03", "2014-04-03", "2015-04-03", "2016-04-03"))
+R_vers = paste0("R 3.", 0:3)
+abline(v=R_dates, col="grey90", lty=2)
+text(R_dates, 11800, R_vers,  pos=4, col="grey50")
+
+# Label lines
+lines(rcpp$date, rcpp$Downloads,  lwd=3,col=3)
+lines(v8$date, v8$Downloads, lwd=2, col=4)
+lines(rjava$date, rjava$Downloads, lwd=2, col=2)
+lines(rpython$date, rpython$Downloads, lwd=2, col=1)
+
+# Add axis
+axis(2, tick=FALSE)
+axis(1, at = c(as.Date("2013-01-01"),as.Date("2014-01-01"),as.Date("2015-01-01"), as.Date("2016-01-01")), 
+     labels=2013:2016, tick=F)
+
+title("The rise of Rcpp", adj=1, 
+      cex.main=1, font.main=2, col.main="black")
+
+# Label lines
+text(as.Date("2013-06-01"), 6, "rPython", col=1, font = 2)
+text(as.Date("2014-02-01"), 3000, "Rcpp", col=3, font = 2)
+text(as.Date("2015-02-01"), 80, "V8", col=4, font = 2)
+text(as.Date("2015-03-01"), 1000, "rJava", col=2, font = 2)
+

code/07-performance_f4.R → code/07-performance_f6.R

@@ -7,23 +7,23 @@ z$expr = factor(z$expr, levels=c("mean(x)",  "mean_c(x)", "com_mean_r(x)", "mean
 z$expr = factor(z$expr, labels=c("Base",  "Rcpp", "Byte compiled R", "Pure R"))
 
 z$time = z$time/10^6
-g = ggplot(z) + 
-  geom_violin(aes(expr, time),position=position_dodge(0.9), bg=get_col(3)) + 
-  scale_y_continuous(limits=c(10^-3, 10^1), expand=c(0, 0), breaks = 10^(-3:1), 
-                     trans="log10", 
+g = ggplot(z) +
+  geom_violin(aes(expr, time),position=position_dodge(0.9), bg=get_col(3)) +
+  scale_y_continuous(limits=c(10^-3, 10^1), expand=c(0, 0), breaks = 10^(-3:1),
+                     trans="log10",
                      labels=c(expression(10^-3),expression(10^-2),
                               expression(10^-1),expression(10^0),expression(10^1)))
 
-g1 = g +  labs(title = "Performance Gains with Rcpp", 
-           x = NULL, y = "Elapsed Time (secs)",colour = NULL, fill = NULL) + 
-  theme(panel.grid.major.y = element_line(colour = "gray90"), 
-        panel.grid.minor = element_line(colour = NA), 
-        panel.grid.major.x = element_line(colour = NA), 
-        plot.title = element_text(size = 12, 
-                                  face = "bold", hjust = 1, vjust = 0), 
-        panel.background = element_rect(fill = NA), 
-        legend.background = element_rect(fill = NA), 
-        legend.position = c(0.93, 0.92), 
+g1 = g +  labs(title = "Performance Gains with Rcpp",
+           x = NULL, y = "Elapsed Time (secs)",colour = NULL, fill = NULL) +
+  theme(panel.grid.major.y = element_line(colour = "gray90"),
+        panel.grid.minor = element_line(colour = NA),
+        panel.grid.major.x = element_line(colour = NA),
+        plot.title = element_text(size = 12,
+                                  face = "bold", hjust = 1, vjust = 0),
+        panel.background = element_rect(fill = NA),
+        legend.background = element_rect(fill = NA),
+        legend.position = c(0.93, 0.92),
         axis.ticks.x = element_line(linetype = "blank"),
         axis.ticks.y = element_line(linetype = "blank"))
-print(g1)
+print(g1)

src/mean_c.cpp → src/mean_cpp.cpp

@@ -2,19 +2,19 @@
 using namespace Rcpp;
 
 // [[Rcpp::export]]
-double mean_c(NumericVector x){
+double mean_cpp(NumericVector x){
   int i;
   int n = x.size();
   double mean = 0;
-  
+
   for(i=0; i<n; i++){
     mean = mean + x[i]/n;
   }
   return mean;
 }
 
 // [[Rcpp::export]]
-NumericVector res_c(NumericVector x, NumericVector y){
+NumericVector res_cpp(NumericVector x, NumericVector y){
   int i;
   int n = x.size();
   NumericVector residuals(n);