1berry/nodejs
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

benchmark: use "confidence" in output of compare.R

Browse Source 
Use the word "confidence" to indicate the confidence level of
the p value so it's easier to understand.
With this change more stars in the output of compare.R means
higher confidence level (lower significance level).

PR-URL: https://github.com/nodejs/node/pull/10737
Refs: https://github.com/nodejs/node/pull/10439
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Andreas Madsen <amwebdk@gmail.com>
This commit is contained in:
Joyee Cheung 2017-01-11 20:16:25 +08:00
parent 2f885e7a43
commit f18b4632b9
2 changed files with 10 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
benchmark/README.md
View File

@ -161,7 +161,7 @@ For analysing the benchmark results use the `compare.R` tool.
```console
$ cat compare-pr-5134.csv | Rscript benchmark/compare.R
improvement significant p.value
improvement confidence p.value
string_decoder/string-decoder.js n=250000 chunk=1024 inlen=1024 encoding=ascii 12.46 % *** 1.165345e-04
string_decoder/string-decoder.js n=250000 chunk=1024 inlen=1024 encoding=base64-ascii 24.70 % *** 1.820615e-15
string_decoder/string-decoder.js n=250000 chunk=1024 inlen=1024 encoding=base64-utf8 23.60 % *** 2.105625e-12
@ -171,7 +171,7 @@ string_decoder/string-decoder.js n=250000 chunk=1024 inlen=128 encoding=ascii
```
In the output, _improvement_ is the relative improvement of the new version,
hopefully this is positive. _significant_ tells if there is enough
hopefully this is positive. _confidence_ tells if there is enough
statistical evidence to validate the _improvement_. If there is enough evidence
then there will be at least one star (`*`), more stars is just better. **However
if there are no stars, then you shouldn't make any conclusions based on the
@ -189,7 +189,7 @@ may require more runs to obtain (can be set with `--runs`).
_For the statistically minded, the R script performs an [independent/unpaired
2-group t-test][t-test], with the null hypothesis that the performance is the
same for both versions. The significant field will show a star if the p-value
same for both versions. The confidence field will show a star if the p-value
is less than `0.05`._
The `compare.R` tool can also produce a box plot by using the `--plot filename`
@ -202,7 +202,7 @@ keep the first line since that contains the header information.
```console
$ cat compare-pr-5134.csv | sed '1p;/encoding=ascii/!d' | Rscript benchmark/compare.R --plot compare-plot.png
improvement significant p.value
improvement confidence p.value
string_decoder/string-decoder.js n=250000 chunk=1024 inlen=1024 encoding=ascii 12.46 % *** 1.165345e-04
string_decoder/string-decoder.js n=250000 chunk=1024 inlen=128 encoding=ascii 6.70 % * 2.928003e-02
string_decoder/string-decoder.js n=250000 chunk=1024 inlen=32 encoding=ascii 7.47 % *** 5.780583e-04

12
benchmark/compare.R
View File

@ -46,7 +46,7 @@ statistics = ddply(dat, "name", function(subdat) {
improvement = sprintf("%.2f %%", ((new.mu - old.mu) / old.mu * 100));
p.value = NA;
significant = 'NA';
confidence = 'NA';
# Check if there is enough data to calulate the calculate the p-value
if (length(old.rate) > 1 && length(new.rate) > 1) {
# Perform a statistics test to see of there actually is a difference in
@ -56,19 +56,19 @@ statistics = ddply(dat, "name", function(subdat) {
# Add user friendly stars to the table. There should be at least one star
# before you can say that there is an improvement.
significant = '';
confidence = '';
if (p.value < 0.001) {
significant = '***';
confidence = '***';
} else if (p.value < 0.01) {
significant = '**';
confidence = '**';
} else if (p.value < 0.05) {
significant = '*';
confidence = '*';
}
}
r = list(
improvement = improvement,
significant = significant,
confidence = confidence,
p.value = p.value
);
return(data.frame(r));