How can I plot benchmark output?

问题

I am learning rbenchmark package to benchmark algorithm and see the performance in R environment. However, when I increased the input, benchmark result are varied one to another. To show how the performance of algorithm for different input, producing line graph or curve is needed. I expect to have one line or curve that show the performance difference of using different number of input. The algorithm I used, works O(n^2) .In resulted plot, X axis show number of observation of input, Y axis shows the run time respectively.How can I make this happen more elegantly by using ggplo2 ? Can anyone give me some idea to generate desired plot ? Any idea please ?

Let's imagine, these are input files :

foo.csv
bar.csv
cat.csv

benchmark result when I used two csv files as an input :

df_2 <- data.frame(
    test=c("s3","s7","s4" ,"s1" ,"s2" ,"s5" ,"s6" ,"s9","s8"),
    replications=c(10,10, 10, 10 ,10 ,10 ,10 ,10 ,10),
    elapsed=c(0.23,  0.28,  0.53 , 0.80 , 4.12 , 8.57 , 8.81 ,20.16 ,24.53),
    relative=c( 1.000 ,  1.217 ,  2.304 ,  3.478 , 17.913 , 37.261 , 38.304 , 87.652 ,106.652),
    user.self=c(0.23, 0.28 , 0.53 , 0.61 , 4.13 , 8.55 , 8.80 ,18.06 ,19.08),
    sys.self=c(0.00, 0.00 ,0.00, 0.00 ,0.00, 0.00 ,0.00 ,0.13, 0.51)
)

This time I used three csv files as an input :

df_3 <- data.frame(
    test=c("s3", "s7" ,"s4", "s1", "s5", "s6","s2", "s9","s8"),
    replications=c(10,10, 10, 10 ,10 ,10 ,10 ,10 ,10),
    elapsed=c( 0.34 , 0.47 , 0.70 , 2.41  ,8.26 , 8.75 , 9.03, 28.78 ,36.56),
    relative=c( 1.000 ,  1.382 ,  2.059  , 7.088 , 24.294 , 25.735 , 26.559  ,84.647 ,107.529),
    user.self=c(0.34 , 0.46  ,0.70 , 1.72 , 8.26 , 8.74  ,9.01, 26.24 ,30.95),
    sys.self=c(0.00 ,0.00 ,0.00, 0.12, 0.00 ,0.00 ,0.00, 0.12 ,0.77)
)

In my desired plot, two line plot or curve must be placed in one grid.

How can I get nice line graph or curve by using above benchmark result ? How can I achieve desired plot that show performance of algorithm in R ? Thanks a lot

回答1:

You can try this (assuming that s1, s2, s3, ... represent different tests, possibly with different n, that you want to compare, with the results df_2 against df_3):

library(reshape2)
df_2 <- melt(df_2, id='test')
df_3 <- melt(df_3, id='test')
df_2$num_input <- 'two_input'
df_3$num_input <- 'three_input'
df <- rbind(df_2, df_3)
library(ggplot2)
ggplot(df, aes(test, value, group=num_input, col=num_input)) + geom_point() + geom_line() + facet_wrap(~variable)

If you want to plot elapsed against test try this:

ggplot(df[df$variable=='elapsed',], aes(test, value, group=num_input, col=num_input)) + geom_point() + geom_line(lwd=2) + ylab('elapsed') +
  theme(text=element_text(size=15))

If you want more readable images, try this:

ggplot(df, aes(test, value, group=num_input, col=num_input)) + geom_point() + geom_line(lwd=2) + facet_wrap(~variable) +
  theme(text=element_text(size=15))

[EDITED] geom_smooth

ggplot(df[df$variable=='elapsed',], aes(test, value, group=num_input, col=num_input)) + 
  geom_point() + geom_smooth(span=0.7, se=FALSE) + ylab('elapsed') +
  theme(text=element_text(size=15))

回答2:

First, we create a grouping variable.

df_2$set <- "set_1"
df_3$set <- "set_2"

Then we create a variable for the number of replications.

df_2$n <- 1:length(df_2$replications)
df_3$n <- 1:length(df_2$replications)

We plot binding df_2 and df_3 by rows, creating a single data frame.

This will create a line plot.

ggplot(rbind(df_2, df_3)) + 
  aes(as.factor(n), elapsed, color = set, group = set) + 
  geom_line()

This will create a smooth line plot, using loess as its method.

ggplot(rbind(df_2, df_3)) + 
  aes(as.factor(n), elapsed, color = set, group = set) + 
  geom_smooth(alpha = 0)

来源：https://stackoverflow.com/questions/41523644/how-can-i-plot-benchmark-output

标签

ggplot2

benchmarking