An exploration of the mlr command-line tool and bar charts

[kostmo]

I've been mulling the idea of representing certain data in the swarm UI as bar charts.
For comparison's sake, I was curious what CLI tools exist today to render bar charts in the terminal.

There is is a tool called miller (mlr) that can output to (self-proclaimed "cheesy") bar charts. It can also compute histograms.

^{Sadly, the version of miller (version 6.0.0) available on Ubuntu 22.04 has a bug in its histogram computation with --auto, so I installed version 6.12.0-dev from source.}

First, computing a histogram of line counts of .hs source files:

$ find src -name '*.hs' -exec wc -l {} \; | mlr --nidx --opprint histogram -f 1 --nbins 8 --hi 2000
bin_lo bin_hi 1_count
0      250    173
250    500    29
500    750    6
750    1000   3
1000   1250   0
1250   1500   1
1500   1750   2
1750   2000   1

Next, appending a bar-chart rendering stage to the mlr command with then:

$ find src -name '*.hs' -exec wc -l {} \; | mlr --nidx --opprint histogram -f 1 --nbins 8 --hi 2000 then bar -w 60 --auto -f 1_count -b ' '
bin_lo bin_hi 1_count
0      250    [0]***********************************************************#[173]
250    500    [0]**********                                                  [173]
500    750    [0]**                                                          [173]
750    1000   [0]*                                                           [173]
1000   1250   [0]#                                                           [173]
1250   1500   [0]#                                                           [173]
1500   1750   [0]#                                                           [173]
1750   2000   [0]#                                                           [173]

The # character represents "out of bounds", given the resolution of the bar increments. We can suppress this indicator:

$ find src -name '*.hs' -exec wc -l {} \; | mlr --nidx --opprint histogram -f 1 --nbins 8 --hi 2000 then bar -w 60 --auto -f 1_count -b ' ' -x ' '

Conclusions

• At least for high-variance data, bar charts alone can be harder to read in the terminal than just plain numbers.
• The bars should be supplemented by their actual numeric values.
• For histograms, it would be nice to distinguish between values "out-of-bounds" because they are zero, vs. because they are small non-zero values.

[kostmo]

I was also looking into gnuplot to generate a histogram of file line counts. gnuplot itself cannot perform binning on the data; some external program is needed to pre-process the data into bins. mlr can do that:

$ find src -name '*.hs' -exec wc -l {} \; | mlr --nidx --otsv histogram -f 1 --nbins 8 --hi 2000 then rename 1_count,count then cut -f bin_hi,count | gnuplot -p -e 'set style data histograms; plot "-" using 2:xticlabels(1) title columnheader fs pattern 2'

To render to file:

$ find src -name '*.hs' -exec wc -l {} \; | mlr --nidx --otsv histogram -f 1 --nbins 8 --hi 2000 then rename 1_count,count then cut -f bin_hi,count | gnuplot -e 'set style data histograms; set terminal png; plot "-" using 2:xticlabels(1) title columnheader fs pattern 2' > plot.png