Often outliers are present in data set.
gnuplot uses the minimum and the maximum of the data for autoscaling.
But any ideas how to make the scaling robust against outliers? I'm thinking about bind and a function that computes quantiles/percentiles.
array DATA[1000]
do for [i=1:1000] {DATA[i]=invnorm(rand(0))}
DATA[42] = 1e7
pl DATA pt 7
pl [][-4:4] DATA pt 7
After a lot of interactive zooming or manual adjusting of the plot range:
The stats command will report quartiles. You could bind a keystroke to rescale on y using the previous result from stats. Using the median +/- 1.5 * (inter-quartile range) would give the same overall range as the default whisker bars on a boxplot.
bind 'S' 'set yrange [ STATS_median_y - 1.5 * (STATS_up_quartile_y-STATS_lo_quartile_y) : STATS_median_y + 1.5 * (STATS_up_quartile_y-STATS_lo_quartile_y) ]; replot'
stats DATA using 1:2
plot DATA using 1:2
pause -1 "Type 'S' in plot window to rescale"
Compare this to
plot DATA using 1:2 with boxplot
2.22.