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

Groups > comp.graphics.apps.gnuplot > #794 > unrolled thread

Calculate R-squared value from gnuplot 'fit' output?

Back to article view | Back to comp.graphics.apps.gnuplot

Contents

  Calculate R-squared value from gnuplot 'fit' output? Mike <mesimms@gmail.com> - 2011-12-28 12:14 -0800
    Re: Calculate R-squared value from gnuplot 'fit' output? Hans-Bernhard Bröker <HBBroeker@t-online.de> - 2011-12-28 22:11 +0100
    Re: Calculate R-squared value from gnuplot 'fit' output? wolfgang <cmat.wolfgang@gmail.com> - 2011-12-29 00:17 -0800
      Re: Calculate R-squared value from gnuplot 'fit' output? Hans-Bernhard Bröker <HBBroeker@t-online.de> - 2011-12-29 12:39 +0100
      Re: Calculate R-squared value from gnuplot 'fit' output? ab <ab.3942@googlemail.com> - 2012-01-02 00:37 -0800
    Re: Calculate R-squared value from gnuplot 'fit' output? benjamin.chen@aztopia.com - 2013-11-04 19:17 -0800
      Re: Calculate R-squared value from gnuplot 'fit' output? Karl <mail.kfr@gmx.net> - 2013-11-05 14:14 +0100
        Re: Calculate R-squared value from gnuplot 'fit' output? ooe@st-andrews.ac.uk - 2014-12-10 16:09 -0800
          Re: Calculate R-squared value from gnuplot 'fit' output? ali.a.abedi@gmail.com - 2015-01-26 11:32 -0800
            Re: Calculate R-squared value from gnuplot 'fit' output? thiagopaes@gmail.com - 2016-07-08 12:58 -0700
              Re: Calculate R-squared value from gnuplot 'fit' output? luis.blacutt@gmail.com - 2016-07-15 08:29 -0700
                Re: Calculate R-squared value from gnuplot 'fit' output? armin@armbrusters.name - 2017-02-21 12:40 -0800
    Re: Calculate R-squared value from gnuplot 'fit' output? henryxparker@gmail.com - 2017-10-13 10:25 -0700
      Re: Calculate R-squared value from gnuplot 'fit' output? Karl Ratzsch <mail.kfr@gmx.net> - 2017-10-13 19:53 +0200
        Re: Calculate R-squared value from gnuplot 'fit' output? ruben safir <ruben@mrbrklyn.com> - 2017-10-24 17:35 -0400
          Re: Calculate R-squared value from gnuplot 'fit' output? Karl Ratzsch <mail.kfr@gmx.net> - 2017-10-25 09:27 +0200

#794 — Calculate R-squared value from gnuplot 'fit' output?

OK, my statistics is perhaps not what it should be. I have a script
make a linear fit of some x-y pairs:

#
# Correlations
#
set xdata
set format x
set term png size 400, 400
f(x) = m*x + b
set output "./corr-00.png"
fit f(x) "2011-10.dat" using 2:3 via m,b
plot "2011-10.dat" using 2:3 t "C1 vs C2" with dots, f(x) title "Model
Fit"

And I get the following output, just showing the end here:

 Iteration 0
 WSSR        : 9404.93           delta(WSSR)/WSSR   : 0
 delta(WSSR) : 0                 limit for stopping : 1e-05
 lambda	  : 7.53716

After 3 iterations the fit converged.
final sum of squares of residuals : 1980.57
rel. change during last iteration : -5.27226e-11

degrees of freedom    (FIT_NDF)                        : 4462
rms of residuals      (FIT_STDFIT) = sqrt(WSSR/ndf)    : 0.66624
variance of residuals (reduced-chisquare) = WSSR/ndf   : 0.443875

Final set of parameters            Asymptotic Standard Error
=======================            ==========================

m               = 0.889071         +/- 0.002949     (0.3317%)
b               = 0.881951         +/- 0.03129      (3.548%)


correlation matrix of the fit parameters:

               m      b
m               1.000
b              -0.948  1.000

Shouldn't I be able to calculate the R-squared value from this data? I
read in wikipedia that

R^2 1 - (SS_err / SS_tot)

and that WSSR is SS_err, as far as I can make out. Can anyone help me
with SS_tot or another way to do it?

Thanks!

[toc] | [next] | [standalone]

#795

On 28.12.2011 21:14, Mike wrote:

> Shouldn't I be able to calculate the R-squared value from this data?

In a word: no.

R^2 doesn't have any useful relation to what 'fit' calculates.

[toc] | [prev] | [next] | [standalone]

#796

On Dec 28, 3:14 pm, Mike <mesi...@gmail.com> wrote:
> OK, my statistics is perhaps not what it should be. I have a script
> make a linear fit of some x-y pairs:
>
> #
> # Correlations
> #
> set xdata
> set format x
> set term png size 400, 400
> f(x) = m*x + b
> set output "./corr-00.png"
> fit f(x) "2011-10.dat" using 2:3 via m,b
> plot "2011-10.dat" using 2:3 t "C1 vs C2" with dots, f(x) title "Model
> Fit"
>
> And I get the following output, just showing the end here:
>
>  Iteration 0
>  WSSR        : 9404.93           delta(WSSR)/WSSR   : 0
>  delta(WSSR) : 0                 limit for stopping : 1e-05
>  lambda   : 7.53716
>
> After 3 iterations the fit converged.
> final sum of squares of residuals : 1980.57
> rel. change during last iteration : -5.27226e-11
>
> degrees of freedom    (FIT_NDF)                        : 4462
> rms of residuals      (FIT_STDFIT) = sqrt(WSSR/ndf)    : 0.66624
> variance of residuals (reduced-chisquare) = WSSR/ndf   : 0.443875
>
> Final set of parameters            Asymptotic Standard Error
> =======================            ==========================
>
> m               = 0.889071         +/- 0.002949     (0.3317%)
> b               = 0.881951         +/- 0.03129      (3.548%)
>
> correlation matrix of the fit parameters:
>
>                m      b
> m               1.000
> b              -0.948  1.000
>
> Shouldn't I be able to calculate the R-squared value from this data? I
> read in wikipedia that
>
> R^2 1 - (SS_err / SS_tot)
>
> and that WSSR is SS_err, as far as I can make out. Can anyone help me
> with SS_tot or another way to do it?
>
> Thanks!

Why does that take 3 iterations? That is just a linear LS problem?
Linear LS regression has an analytical solution, i.e. with normal
equations.
Solution should be either intelligent starting values or one iteration
from
dummy starting point?
Wolfgang

[toc] | [prev] | [next] | [standalone]

#797

On 29.12.2011 09:17, wolfgang wrote:

> Why does that take 3 iterations? That is just a linear LS problem?

Because 'fit' is not linear regression.

Generic tools generally don't work as smoothly as special-purpose ones.

[toc] | [prev] | [next] | [standalone]

#802

>
> Why does that take 3 iterations? That is just a linear LS problem?
> Linear LS regression has an analytical solution, i.e. with normal
> equations.
> Solution should be either intelligent starting values or one iteration
> from

Hi,

You can force the fit  to converge in one iteration by using
FIT_START_LAMBDA=1.e-15
FIT_MAXITER=1
This is only suited for a linear model.

A better way is the stats command in gnuplot 4.5.
The correlation coefficient  r is calculated directly.

[toc] | [prev] | [next] | [standalone]

#2189

El miércoles, 28 de diciembre de 2011 15:14:18 UTC-5, Mike  escribió:
> OK, my statistics is perhaps not what it should be. I have a script
> 
> make a linear fit of some x-y pairs:
> 
> 
> 
> #
> 
> # Correlations
> 
> #
> 
> set xdata
> 
> set format x
> 
> set term png size 400, 400
> 
> f(x) = m*x + b
> 
> set output "./corr-00.png"
> 
> fit f(x) "2011-10.dat" using 2:3 via m,b
> 
> plot "2011-10.dat" using 2:3 t "C1 vs C2" with dots, f(x) title "Model
> 
> Fit"
> 
> 
> 
> And I get the following output, just showing the end here:
> 
> 
> 
>  Iteration 0
> 
>  WSSR        : 9404.93           delta(WSSR)/WSSR   : 0
> 
>  delta(WSSR) : 0                 limit for stopping : 1e-05
> 
>  lambda	  : 7.53716
> 
> 
> 
> After 3 iterations the fit converged.
> 
> final sum of squares of residuals : 1980.57
> 
> rel. change during last iteration : -5.27226e-11
> 
> 
> 
> degrees of freedom    (FIT_NDF)                        : 4462
> 
> rms of residuals      (FIT_STDFIT) = sqrt(WSSR/ndf)    : 0.66624
> 
> variance of residuals (reduced-chisquare) = WSSR/ndf   : 0.443875
> 
> 
> 
> Final set of parameters            Asymptotic Standard Error
> 
> =======================            ==========================
> 
> 
> 
> m               = 0.889071         +/- 0.002949     (0.3317%)
> 
> b               = 0.881951         +/- 0.03129      (3.548%)
> 
> 
> 
> 
> 
> correlation matrix of the fit parameters:
> 
> 
> 
>                m      b
> 
> m               1.000
> 
> b              -0.948  1.000
> 
> 
> 
> Shouldn't I be able to calculate the R-squared value from this data? I
> 
> read in wikipedia that
> 
> 
> 
> R^2 1 - (SS_err / SS_tot)
> 
> 
> 
> and that WSSR is SS_err, as far as I can make out. Can anyone help me
> 
> with SS_tot or another way to do it?
> 
> 
> 
> Thanks!

Try using this instead:
stats 'file.dat' using 1:2 name "A"

In should print out something like this:
* FILE: 
  Records:      21
  Out of range:  0
  Invalid:       0
  Blank:         1
  Data Blocks:   1

* COLUMNS:
  Mean:        500.0000             7.5762
  Std Dev:     302.7650             3.9789
  Sum:       10500.0000           159.1000
  Sum Sq.:  7.17500e+06          1537.8300

  Minimum:       0.0000 [ 0]        1.0000 [ 0]
  Maximum:    1000.0000 [20]       14.1000 [20]
  Quartile:    250.0000             4.5000
  Median:      500.0000             7.2000
  Quartile:    750.0000            11.1000

  Linear Model: y = 0.01311 x + 1.019
  Correlation:  r = 0.9979
  Sum xy:       1.048e+05

[toc] | [prev] | [next] | [standalone]

#2190

Am 05.11.2013 04:17, schrieb benjamin.chen@aztopia.com:
> El miércoles, 28 de diciembre de 2011 15:14:18 UTC-5, Mike  escribió:


It´s useless, most likely, to answer to a newspost question that is two
years old. Bonus points for the effort, but your comment will probably
be lost.

 K

[toc] | [prev] | [next] | [standalone]

#2666

On Tuesday, November 5, 2013 1:14:06 PM UTC, Karl wrote:
> Am 05.11.2013 04:17, schrieb benjamin.chen@aztopia.com:
> > El miércoles, 28 de diciembre de 2011 15:14:18 UTC-5, Mike  escribió:
> 
> 
> It´s useless, most likely, to answer to a newspost question that is two
> years old. Bonus points for the effort, but your comment will probably
> be lost.
> 
>  K

It's not useless in this era, somebody may be looking for this help 5 years from now.

[toc] | [prev] | [next] | [standalone]

#2732

On Wednesday, December 10, 2014 at 7:09:29 PM UTC-5, o...@st-andrews.ac.uk wrote:
> On Tuesday, November 5, 2013 1:14:06 PM UTC, Karl wrote:
> > Am 05.11.2013 04:17, schrieb benjamin.chen@aztopia.com:
> > > El miércoles, 28 de diciembre de 2011 15:14:18 UTC-5, Mike  escribió:
> > 
> > 
> > It´s useless, most likely, to answer to a newspost question that is two
> > years old. Bonus points for the effort, but your comment will probably
> > be lost.
> > 
> >  K
> 
> It's not useless in this era, somebody may be looking for this help 5 years from now.

Yes I used it 2 years later. Google indexes all of this stuffs and it was a hit for my query.

[toc] | [prev] | [next] | [standalone]

#3368

I used it NOW =)

[toc] | [prev] | [next] | [standalone]

#3371

On Friday, July 8, 2016 at 4:58:20 PM UTC-3, thiag...@gmail.com wrote:
> I used it NOW =)

Not exactly five years, but it certainly answered my question

[toc] | [prev] | [next] | [standalone]

#3564

On Friday, July 15, 2016 at 11:29:22 AM UTC-4, luis.b...@gmail.com wrote:
> On Friday, July 8, 2016 at 4:58:20 PM UTC-3, thiag...@gmail.com wrote:
> > I used it NOW =)
> 
> Not exactly five years, but it certainly answered my question

More than 5 years by now and it's still helpful!

[toc] | [prev] | [next] | [standalone]

#3770

Still relevant!

[toc] | [prev] | [next] | [standalone]

#3771

Am 13.10.2017 um 19:25 schrieb henryxparker@gmail.com:
> Still relevant!
> 

This is a newsgroup.

1. Be polite, and sign your message, and
2. cite the original post, because most people don't read this on
googles stolen archive pages, but in their newsreader, and delete
old messages after a few months.

  Karl

[toc] | [prev] | [next] | [standalone]

#3797

On 10/13/2017 01:53 PM, Karl Ratzsch wrote:
> Am 13.10.2017 um 19:25 schrieb henryxparker@gmail.com:
>> Still relevant!
>>
> This is a newsgroup.
> 

First of all, your an idiot.

> 1. Be polite, and sign your message, and

Be polite and stop trying to control others behavior.  Don't post
messages like this.  If you have nothing constructive to say, say nothing.


> 2. cite the original post, 

If nobody knows what he refers to, that is his problem.  If you want to
ask him wtf he is refering to, just ask.

> because most people don't read this on
> googles stolen archive pages, 

That archive, which actually originated with Dejanews before you grew
out of diapers, is in no way stolen and is blessing for the public.

> but in their newsreader, and delete
> old messages after a few months.
>

And if not, you your eyes they are second class humans unfit to stand
with German Supermen like yourself?


Go find some therapy, smoke some grass, or get layed,  Do anything more
than troll innocent usenet posters.

[toc] | [prev] | [next] | [standalone]

#3804

Am 24.10.2017 um 23:35 schrieb ruben safir:

> messages like this.  If you have nothing constructive to say, say nothing.

Word.

[toc] | [prev] | [standalone]

Back to top | Article view | comp.graphics.apps.gnuplot

csiph-web