Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.graphics.apps.gnuplot > #3568

Re: Strange thing about logscale after a fit ?

Show all headers | View raw

Vincent Gouret wrote:

> Hello all,
> 
> I have a strange result for plotting data in logscale. The file I use
> contains this:
> --------------------
> temps_gauss.txt:
> 
> 5	100000	4
> 6	69444	4
> 7	51020	4
> 9	30863	5
> 10	24999	5
> 12	17360	6
> 13	14791	7
> 16	9764	9
> 18	7714	11
> 21	5667	15
> 25	3998	22
> 29	2971	31
> 34	2161	47
> 40	1561	70
> 47	1130	110
> 55	825	157
> 64	609	232
> 75	443	361
> 88	321	552
> 103	234	822
> 120	172	1281
> 141	124	2031
> 165	90	3148
> 193	65	4941
> 227	46	7883
> 265	33	12483
> 311	23	19784
> 364	16	31094
> 427	11	50103
> 501	7	80055
> ----------------------
> The first colum is the size of a linear problem, the second is the number
> of random problems of this size and the last is the mean execution time in
> us. The linear problem is solve using the Gauss algorithm. I plot this
> result in Gnuplot with: plot 'temps_gauss.txt' u 1:3 w p It work fine. If
> I want to see the same plot with logscale, I do this: set logscale
> and when I redo the plot, all goes fine. After that, I do a fit like that:
> f(x)=alpha*x**3
> fit f(x) 'temps_gauss.txt' u 1:3 via alpha
> This works well and give me a value for alpha=6.4e-4, fine. Then I plot
> the datas and the model, and at this time it does not work: plot
> 'temps_gauss.txt' u 1:3 w p, f(x) w l This give me the error:
>  color axis has cb coord of 0; must be above 0 for log scale!
> If I stop gnuplot, and relaunch it, and type:
> f(x)=alpha*x**3
> set logscale
> alpha=6.4e-4
> plot 'temps_gauss.txt' u 1:3 w p, f(x) w l
> then all goes fine ??? What is the problem ? I use gnuplot V5 patchlevel
> 4, under Windows 8.1 or V5 patchlevel 3 under Lubuntu 16.04.2, the same...

The error occurs because 

- the command "set logscale" applies to all axes, not just x and y.  
- the default range for all axes in the absence of any input is [-10:10]
- if you actually use an axis in the plot, the program checks that the
  range is reasonable, which in the case of log scale means 
  adjusting the default it to be > 0 if necessary.
- unfortunately (here is the bug) the color axis range is not adjusted
  for log scale because your plot does not actually use it.  So the 
  default range is incompatible with log scale and you get an error.
  I do not know exactly how long this bug has been there, but I will file
  a bug tracker item so that it gets fixed in the next gnuplot release.

To avoid the error, any one of the following is sufficient

- Only set log scale for x and y:
    set log xy
- Clear log scale for the color axis:
    unset log cb
- Set the range of the color axis to be >0
    set cbrange [1:*]


	Ethan

Back to comp.graphics.apps.gnuplot | Previous | Next — Previous in thread | Next in thread | Find similar

Thread

Strange thing about logscale after a fit ? Vincent Gouret <vincentgouret44@gmail.com> - 2017-03-08 10:07 -0800
  Re: Strange thing about logscale after a fit ? Ethan A Merritt <sfeam@users.sourceforge.net> - 2017-03-08 11:01 -0800
    Re: Strange thing about logscale after a fit ? Vincent Gouret <vincentgouret44@gmail.com> - 2017-03-08 11:16 -0800

csiph-web