Customising the analysis workflow

While the configuration of the execution workflow is completely straight-forward, the analysis workflow is a bit more complicated.

For its usage of Gnuplot and Gnu R, it automatically writes scripts - which you, the user, might want a little different. Also, the refinement step (selecting values out of filed to plot) can be configured.

Here are a few short tips on how to customise certain things of the analysis:

Custom selectors

StoSim lets you select some values from the columns you want to analyse, e.g. the maximum value or simply the last one. You can also write a custom selector and tell StoSim to use it.

Warning

Custom selectors are not implemented yet, but high on the TODO-list!

Custom plots

The possibilities of plot-making in StoSim are pretty interesting, but of course one can do a lot more with Gnuplot. It is easy to link your own Gnuplot script into the workflow.

There is an additional configuration setting per figure, called custom-script. Put here the path to your custom script, relative to the simulation folder (that is the first argument to StoSim), e.g.:

[figure1]
name: my_custom_plot
y-range: [0:50]
custom-script: my_scripts/figure1.gnu
plot1: _name:fig1_plot1, _ycol:2, _type:line

Note

Graphical settings like y-range will not have any effect if you use a custom script.

Note

The plot<i> directives are helping you gather data, so you’ll need one for every set of data you want to plot.

Your script can make use of the work StoSim has done on selecting files, columns and selected values in them. For each plot, your script can access these in a file called all.dat, to be found in a directory named after the _name attribute of the plot.

As a reference, here is a Gnuplot file that StoSim creates itself to make line plots (from the tutorial Using subsimulations):

set terminal postscript  eps color dashed lw 6 rounded 22;
set output 'simulation2_payoff.eps';
set yrange [0:7];
set xlabel 'iteration';
set ylabel 'payoff';
set key top left spacing 1;
plot 'non-learners_in_coop/all.dat' smooth unique  title 'non-learners_in_coop' lt 1,'learners_in_coop/all.dat' smooth unique  title 'learners_in_coop' lt 2,'non-learners_in_non-coop/all.dat' smooth unique  title 'non-learners_in_non-coop' lt 3,'learners_in_non-coop/all.dat' smooth unique  title 'learners_in_non-coop' lt 4

Note

You can see StoSims own Gnuplot script and all the data files it creates for the analysis by simply adding -k to the call (e.g. ./stosim . --plots -k). This will keep StoSim from deleting its temporary files. There should now be a directory called tmp_plotter, where you’ll find a file called plot.gnu and a directory with collected data for the last generated plot. That might be a good starting point.

Custom tests

It is also possible to link custom Gnu R scripts into the analysis workflow. It works exactly like in the plotting case above. You add a custom-script setting to your test-description. Like so:

[ttest1]
name: condition1_yes_vs_no
custom-script: my_scripts/ttest1.r
set1: _name:yes, _col:2, condition1:yes
set2: _name:no, _col:2, condition1:no

Each dataset will be in a file called [set-name].dat, so in this example, you can expect yes.dat and no.dat. The Gnu R scripts StoSim uses are pretty trivial, but here is an example nonetheless:

yes <- read.table('yes.dat')
no <- read.table('no.dat')
t.test(yes, no)

Note

The -k option described above also works with tests. The directory to look in is called tmp_tester.