The barchart package
This package provides a program that generates bar charts from CSV files and has special support for visualising the benchmark results of the criterion and progression tools. It comes with an associated API that can be used to generate custom bar charts in case the charts generated by the command-line program are insufficient for your needs.
[Skip to ReadMe]
|Versions||0.1, 0.1.1, 0.1.1.1|
|Change log||None available|
|Dependencies||base (>=3 && <5), cmdargs (>=0.4 && <0.7), csv (>=0.1 && <1.2), diagrams (==0.2.*), filepath (>=1.1 && <1.3) [details]|
|Uploaded||Fri May 6 17:42:42 UTC 2011 by SebastianFischer|
|Downloads||598 total (33 in last 30 days)|
|Status||Docs uploaded by user|
Build status unknown [no reports yet]
- barchart-0.1.1.1.tar.gz [browse] (Cabal source package)
- Package description (included in the package)
For package maintainers and hackage trustees
Readme for barchart-0.1.1.1
barchart is a command-line program with associated Haskell library for generating bar charts from CSV files. It has special support for creating charts from data generated by the Haskell benchmarking tools criterion and progression. barchart can create PNG, SVG, PDF, and PS files using the Haskell Diagrams library for rendering and, hence, depends on a Haskell binding to Cairo which is part of Gtk2Hs.
cabal install barchart
Bar charts can be created from CSV files using the
command-line utility. For example, if you want to track how many hours
you practice playing the guitar on each day of the week, you can
create a file
guitar.csv with the following contents:
Mon,1.2 Tue,0.3 Wed,2.1 Thu,0.9 Fri,1.1 Sat,3.2 Sun,3.1
# barchart --title="Practice hours" guitar.csv
creates a file
guitar.png which looks like this:
Each bar is labeled with a weekday and has an associated practice
hour. The chart is scaled automatically such that the largest bar
spans the (configurable, see below) height of the chart. The
flag passed to
barchart in the above call is optional. If you do not
supply one, then barchart uses the basename of the CSV file as title
of the chart.
If you want to track practice hours over multiple weeks, you can create a file with a mean, minimal, and maximal values for each day of the week:
Mon,1.2,0.9,1.7 Tue,0.3,0.1,0.5 Wed,2.1,1.5,2.5 Thu,0.9,0.4,1.0 Fri,1.1,1.0,1.2 Sat,3.2,1.5,5.2 Sun,3.1,2.3,4.2
barchart in the
# barchart intervals --title="Mean practice hours" guitar-mean.csv
produces the following bar chart:
In this chart, each bar represents the mean practice hour for a day of the week and the minimum and maximum values are depicted with intervals on the right edge of a bar.
If you want to compare your practice hours for each day of the week and split it by months, you can create a CSV file like this:
Mon,1.2,2.1,1.7 Tue,0.6,0.7,0.8 Wed,2.1,1.2,2.5 Thu,0.9,1.5,1.7 Fri,1.1,1.3,0.7 Sat,3.2,1.7,4.3 Sun,3.1,3.2,2.1
We can use
barchart (in the default mode)
# barchart --title="Practice hours per month" --division="Jan Feb Mar" guitar-months.csv
to create the following diagram:
Each bar is divided into different blocks which all have an associated
amount of practice hours. Green blocks represent practice hours in
January, red blocks in February, and blue blocks represent practice
hours in March. The block labels are given to
barchart via the
--division flag. You can also draw multiple blocks per bar in
interval mode but then three values (mean,min,max) are used
for each block. Hence, if you want to depict mean practice times with
deviations for January, February, and March, you must create a CSV
file where each day of the week is followed by nine practice times.
barchart program can be configured using command-line flags. We
can use the
--help flag to print a summary:
# barchart --help Bar Chart 0.1 barchart [blocks] [FLAG] [FILE] barchart intervals [FLAG] [FILE] barchart criterion [FLAG] [FILE] --summary Show benchmark summary (default) -s --summary-comparison Compare different benchmark summaries -b --benchmark-comparison Compare different benchmarks barchart progression [FLAG] [FILE] -s --summary-comparison Breakdown chart by benchmark summary (default) -b --benchmark-comparison Breakdown chart by benchmarks Common flags: -? --help[=FORMAT] Show usage information (optional format) -V --version Show version information -v --verbose Higher verbosity -q --quiet Lower verbosity -o --out-file=FILE Name of generated file -t --title=STRING Title of bar chart -x --xlabel=STRING Label of x axis -y --ylabel=STRING Label of y axis -g --guess-file-type Guess output file type by name (default) --png Generate .png file --svg Generate .svg file --pdf Generate .pdf file --ps Generate .ps file -d --division=STRINGS Labels separated by whitespace -c --colors=STRINGS Color names separated by whitespace -w --width=NUM Width of generated bar chart (default=600) -h --height=NUM Height of generated bar chart (default=300) -l --label-size=NUM Font size used for labels (default=12) -W --bar-width=FLOAT Bar width between 0 and 1 (default=0.3)
barchart can be run in different modes. We have already seen the
blocks) mode and the
intervals mode. The
progression modes are described below. Most command-line flags are
self explanatory. Apart from what we have seen in the example above,
the following options are particularly interesting:
--ylabellabel the axis of the coordinate system.
--colorschange the colors of the different blocks of a bar. You can use all color names listed in the SVG 1.1 Specification. If you specify fewer colors than there are blocks, then colors are reused in a cyclic fashion. The default value for this argument is
--colors="seagreen firebrick midnightblue".
--heightspecify the dimensions of the generated chart. The generated picture is a little larger because of the title and bar labels. If you want to draw a chart with many bars, you should increase the width compared to the height or tweak the bar width.
--bar-widthis a value between 0.0 and 1.0 hat specifies how thick the bars are compared to the bar distance. With a value of
1.0the bars are drawn directly next to each other, a value of 0.0 draws bars that are actually lines.
barchart command-line program is insufficient for your needs,
you can use the provided API in your own Haskell applications to
generate custom bar charts.
criterion is a Haskell tool for statistically robust benchmarking that can generate graphs which, for example, depict the densities of execution times. criterion can also generate a CSV file summarising all benchmarks but does not provide means to visualise these summaries.
We can use
criterion mode to draw graphs based on the
summary files generated by criterion. To demonstrate the criterion
mode, we write a small Haskell program that benchmarks a simple
definition of the factorial function:
import Criterion.Main main = defaultMain [bgroup "fac" [bench (show n) (nf product [1..n]) | n <- ns]] where ns = [k*10^4 | k <- [1,2,3]] :: [Integer]
We can compile this program, generate a benchmark summary using criterion, and visualise it using barchart as follows:
# ghc -O2 --make factorial # ./factorial --summary=factorial.csv # barchart criterion factorial.csv
These three calls generate a bar chart with one bar for each benchmark whose size is proportional to the corresponding run time.
We can now modify the program to use an explicitly recursive definition of the fibonacci function to see whether this affects the run times.
fac 0 = 1 fac n = n * fac (n-1)
After generating another summary file
factorial2.csv we could
generate another bar chart to visualise it and view both charts side
by side to compare the run times of the different implementations of
the fibonacci function. However, instead of generating two different
charts we can also generate a single chart that combines information
from multiple benchmark summaries.
# barchart criterion --summary-comparison factorial.csv factorial2.csv
We can pass as many summary files as we like and barchart will draw a bar for each summary file with blocks representing the different benchmarks.
We can see clearly that the original implementation using
faster than the explicitly recursive definition of the factorial
function. Instead of drawing different bars for different summaries,
barchart can also draw different bars for the different benchmarks
with blocks for the summaries.
This is less useful for comparing different implementations but useful to compare the different benchmarks using all implementations at once.
progression is an alternative tool to depict criterion's benchmark
results. It generates charts similar to the summary comparisons
discussed above but draws bars for different benchmarks next to each
other rather than as boxes on top of each other in a single
bar. Drawing the results of all benchmarks for a specific
implementation on top of each other lets you see immediately which
implementation is the fastest in total because the total height of
the bar is the sum of all benchmarks for the corresponding
implementation. However, progression provides features that barchart
does not provide such as normalising run times and show them as
percentages. Hence, the barchart program has a
progression mode to
generate bar charts from CSV files generated by progression.
We can alter the factorial program as follows to generate progression's data:
import Criterion import Progression.Main main = defaultMain (bgroup "fac" [bench (show n) (nf fac n) | n <- ns]) where ns = [k*10^4 | k <- [1,2,3]] :: [Integer] fac 0 = 1 fac n = n * fac (n-1)
We do not need to change the fibonacci function itself because we won't execute benchmarks using progression. We only generate a comparison of the previous benchmarks:
# mv factorial.csv bench-factorial.csv # mv factorial2.csv bench-factorial2.csv # ./factorial --mode=graph --compare=factorial,factorial2
progression wants the summary files to have specific names, hence, we
rename them appropriately before generating a graph. The progression
tool generates two files
plot.png. Here, we are not
interested in the generated PNG file but pass the CSV file to the
barchart program in
# barchart progression --division="product recursive" plot.csv
As the CSV file generated by progression does not include information
of the names of the summary files, we provide labels using the
--division flag. The generated bar chart loooks like this:
The run times of the first implementation are normalised to 100 % and
the run times of the second implementation are depicted relative to
the first. Like in
criterion mode we can also flip the axis in
# barchart progression --benchmark-comparison --division="product recursive" plot.csv
Now the first benchmark is normalised to 100 % in each bar so this diagram is of questionable use.
Currently, barchart does not support negative quantities and draws weird bars if you use them anyway. Also, the legend for blocks is sometimes drawn suboptimally. I guess I made a mistake but I cannot find it.
For questions or feedback email Sebastian Fischer.