#211 Cannot get histogram plot to work with in-line function using "reasonable" guess at how
Using version 5.0 patch level 1.0.
I'm trying to get a histogram of an In-line function, but cannot get it to work. Instead, gnuplot does strange stuff. Maybe I'm using it wrong, it seems you must put the data into a file first.
For example, I did this in a cygwin window (simple Bates distribution with n=3, for 20000 samples):
perl -e 'for(1..20000){print ((rand()+rand()+rand())/3.0,"\n")}' > data.txt
Well, okay, actually I needed to use "\r\n" instead of "\n" on Windows. Whatever. Then, in gnuplot:
minx=0.15; maxx=0.85; nsamp=20000.0; nbin=100.0; bin(v)=(v<minx?minx-(maxx-minx)*1.5/nbin:v>maxx?maxx+(maxx-minx)*1.5/nbin:int((v-minx)*(maxx-minx)*nbin+0.5)/nbin/(maxx-minx)+minx); set autoscale xy; \ plot [minx-3*(maxx-minx)/nbin:maxx+3*(maxx-minx)/nbin] "data.txt" using (bin($1)):(1.0/nsamp) smooth frequency with boxes
It gave me exactly the sort of plot I want. But I don't want to have to put the data into a file, especially when all I need to do is check a random distribution. I want to specify the function inline.
But if I try to do the same thing with an in-line function, it simply will not work correctly. Given that the "frequency" style requires x values to count and y values to weigh them with, you would think that going parametric would be required. It sounds reasonable -- let t vary over 20000 samples, but compute each x as the random value (into a bin as before) and let y just be the per-sample weighting (as before). So I've tried that:
minx=0.15; maxx=0.85; nsamp=20000.0; nbin=100.0; bin(v)=(v<minx?minx-(maxx-minx)*1.5/nbin:v>maxx?maxx+(maxx-minx)*1.5/nbin:int((v-minx)*(maxx-minx)*nbin+0.5)/nbin/(maxx-minx)+minx); set autoscale xy; set isosamples nsamp; set parametric; \ plot [minx-3(maxx-minx)/nbin:maxx+3(maxx-minx)/nbin] (bin( (rand(0)+rand(0)+rand(0))/3.0 )),(1.0/nsamp) smooth frequency with boxes
This only shows which x values showed up at all, and not how many times. That is, each box had the same 1.0/nsamp top value, not the count of how many x values fell into that bin. Seems like a defect to me.
It also issues a warning that does not make sense, given that it has boxes for any bin where an x value occurred, which should be stretching vertically from 0 up to 0.00050 along the y axis, but it instead thinks they have no vertical height and tries to spread them across a minuscule range:
Warning: empty y range [5e-005:5e-005], adjusting to [4.95e-005:5.05e-005]
What's up with that?
Too complicated for me to follow.
Can you reduce this to a simpler demonstration?
Anyhow regardless of anything else I don't think isosamples=20000 will ever be reasonable, and I seriously doubt that "set parametric" is compatible with any of your other plot options. That's just not what it's for.
Ticket moved from /p/gnuplot/bugs/1878/
Can't be converted: