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


Groups > comp.soft-sys.math.mathematica > #15740 > unrolled thread

FindFit bug

Started byIgor <igor.volkov@gmail.com>
First post2011-01-31 08:23 +0000
Last post2011-02-01 11:54 +0000
Articles 5 — 5 participants

Back to article view | Back to comp.soft-sys.math.mathematica


Contents

  FindFit bug Igor <igor.volkov@gmail.com> - 2011-01-31 08:23 +0000
    Re: FindFit bug "Sjoerd C. de Vries" <sjoerd.c.devries@gmail.com> - 2011-02-01 11:52 +0000
    Re: FindFit bug Gary Wardall <gwardall@gmail.com> - 2011-02-01 11:54 +0000
    Re: FindFit bug "Kevin J. McCann" <Kevin.McCann@umbc.edu> - 2011-02-01 11:52 +0000
    Re: FindFit bug Albert Retey <awnl@gmx-topmail.de> - 2011-02-01 11:54 +0000

#15740 — FindFit bug

FromIgor <igor.volkov@gmail.com>
Date2011-01-31 08:23 +0000
SubjectFindFit bug
Message-ID<ii5rhr$fh4$1@smc.vnet.net>
Hello,
It seems that FindFit cannot fit a sine function.
It produces no warnings and gives a totally
wrong answer:

dat = Table[{t, 3 Sin[3 t + 1]}, {t, -3, 3, 0.1}];

fit = FindFit[dat, a Sin[w t + f], {a, w, f}, t]

Show[ListPlot[dat], Plot[a Sin[w t + f] /. fit, {t, -3, 3}]]

Output: {a -> 0.599211, w -> 1.51494, f -> 3.80421}

At the same time it fits Sin[3t+1] just fine.

[toc] | [next] | [standalone]


#15811

From"Sjoerd C. de Vries" <sjoerd.c.devries@gmail.com>
Date2011-02-01 11:52 +0000
Message-ID<ii8s6e$2lo$1@smc.vnet.net>
In reply to#15740
This is not a bug.

As the documentation of FindFit clearly states: "In the nonlinear
case, it finds in general only a locally optimal fit."

So, Mathematica has found a local optimum but not a global one due to
its starting values of the parameters. Try providing some starting
parameters that are in the right neighbourhod and you'll see it finds
a much better fit:

fit = FindFit[dat, a Sin[w t + f], {{a, 3.3}, {w, 2.5}, {f, 1.5}}, t]
{a -> 3., w -> 3., f -> 1.}

Another option would be to experiment with different methods:

In[10]:= FindFit[dat, a Sin[w t + f], {a, w, f}, t,
   Method -> #] & /@ {"ConjugateGradient", "Gradient",
  "LevenbergMarquardt", "Newton", "NMinimize", "QuasiNewton"}


Out[10]= {{a -> 8.68943, w -> 0.0199648,
  f -> 0.00865477}, {a -> 2.54082, w -> 0.0678487,
  f -> 0.0294072}, {a -> 0.599211, w -> 1.51494,
  f -> 3.80421}, {a -> 53.7406, w -> 0.00322877,
  f -> 0.00139811}, {a -> -3., w -> -3., f -> -1.}, {a -> 24.9028,
  w -> 0.00696196, f -> 0.00301903}}

In this case NMinimize seems to be the best choice.

Cheers -- Sjoerd

On Jan 31, 9:23 am, Igor <igor.vol...@gmail.com> wrote:
> Hello,
> It seems that FindFit cannot fit a sine function.
> It produces no warnings and gives a totally
> wrong answer:
>
> dat = Table[{t, 3 Sin[3 t + 1]}, {t, -3, 3, 0.1}];
>
> fit = FindFit[dat, a Sin[w t + f], {a, w, f}, t]
>
> Show[ListPlot[dat], Plot[a Sin[w t + f] /. fit, {t, -3, 3}]]
>
> Output: {a -> 0.599211, w -> 1.51494, f -> 3.80421}
>
> At the same time it fits Sin[3t+1] just fine.

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


#15829

FromGary Wardall <gwardall@gmail.com>
Date2011-02-01 11:54 +0000
Message-ID<ii8s9j$2oc$1@smc.vnet.net>
In reply to#15740
On Jan 31, 2:23 am, Igor <igor.vol...@gmail.com> wrote:
> Hello,
> It seems that FindFit cannot fit a sine function.
> It produces no warnings and gives a totally
> wrong answer:
>
> dat = Table[{t, 3 Sin[3 t + 1]}, {t, -3, 3, 0.1}];
>
> fit = FindFit[dat, a Sin[w t + f], {a, w, f}, t]
>
> Show[ListPlot[dat], Plot[a Sin[w t + f] /. fit, {t, -3, 3}]]
>
> Output: {a -> 0.599211, w -> 1.51494, f -> 3.80421}
>
> At the same time it fits Sin[3t+1] just fine.

Numerical procedures are like little girls. =93When they are good they
are very very good but when they are bad they are very very bad=94. In
other words when ever you use a procedure that has a numerical
component you run the risk of not getting the correct answer. In
Mathematica 7=92  look at ?NSolve "Possible Issues"

Try their example:

poly = Expand[Product[x - i, {i, 30}]];
NSolve[poly == 0, x]
Solve[poly == 0, x]


FindFit is a  numerial procedure also and when used use with caution,

Gary Wardall

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


#15843

From"Kevin J. McCann" <Kevin.McCann@umbc.edu>
Date2011-02-01 11:52 +0000
Message-ID<ii8s5o$2l3$1@smc.vnet.net>
In reply to#15740
The problem is that for a nonlinear fit there may be local minima. You 
need to give FindFit a reasonably good initial guess. For example

fit = FindFit[dat, a Sin[w t + f], {{a, 2}, w, f}, t]

starts with a at 2, and the result gives the correct answer. I usually 
use Manipulate to get a good first guess on the parameters for a 1d 
function like this.

Kevin

On 1/31/2011 3:23 AM, Igor wrote:
> Hello,
> It seems that FindFit cannot fit a sine function.
> It produces no warnings and gives a totally
> wrong answer:
>
> dat = Table[{t, 3 Sin[3 t + 1]}, {t, -3, 3, 0.1}];
>
> fit = FindFit[dat, a Sin[w t + f], {a, w, f}, t]
>
> Show[ListPlot[dat], Plot[a Sin[w t + f] /. fit, {t, -3, 3}]]
>
> Output: {a ->  0.599211, w ->  1.51494, f ->  3.80421}
>
> At the same time it fits Sin[3t+1] just fine.
>

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


#15868

FromAlbert Retey <awnl@gmx-topmail.de>
Date2011-02-01 11:54 +0000
Message-ID<ii8s9u$2ok$1@smc.vnet.net>
In reply to#15740
Hi,

> It seems that FindFit cannot fit a sine function.
> It produces no warnings and gives a totally
> wrong answer:
> 
> dat = Table[{t, 3 Sin[3 t + 1]}, {t, -3, 3, 0.1}];
> 
> fit = FindFit[dat, a Sin[w t + f], {a, w, f}, t]
> 
> Show[ListPlot[dat], Plot[a Sin[w t + f] /. fit, {t, -3, 3}]]
> 
> Output: {a -> 0.599211, w -> 1.51494, f -> 3.80421}
> 
> At the same time it fits Sin[3t+1] just fine.

Which version are you using? For me it gives a warning with version 8
but not with version 7. In both cases using Method "NMinimize" yields a
correct result:

fit = FindFit[dat, a Sin[w t + f], {a, w, f}, t,
  Method -> "NMinimize"]

It is interesting to see that version 8 finds the parameters with
negative signs when no additional constraints are given (which is of
course also a correct result and you can enforce the positive signs with
an additional constraint)...

hth,

albert

[toc] | [prev] | [standalone]


Back to top | Article view | comp.soft-sys.math.mathematica


csiph-web