| Re: CURVEFIT for multiple datasets [message #58442] |
Fri, 01 February 2008 01:14 |
Wox
Messages: 184
Registered: August 2006
|
Senior Member |
|
|
On Thu, 31 Jan 2008 04:55:11 -0800 (PST), chloesharrocks@gmail.com
wrote:
<snip>
> How can I extend this for the case when I have say 2data sets, ie:
> x=[[indgen(10)], [indgen(10)]]
> y=[[12.0, 11.0, 10.2, 9.4, 8.7, 8.1, 7.5, 6.9, 6.5, 6.1], [11.5, 10.8,
> 10.3, 9.5, 8.6, 7.8, 7.5, 6.4, 6.4, 5.9]
> and I want to plot an exponential curve through all of this data?
Euhm, I must be missing something here, but what's wrong with just
doing it. You are trying to fit 1 curve which fits best to all
datasets right? I.e. 3 parameters and not 3n parameters?
PRO CURVE_FITTING
x=[findgen(10),findgen(10)]
y=[12.0, 11.0, 10.2, 9.4, 8.7, 8.1, 7.5, 6.9, 6.5, 6.1, 11.5, 10.8,$
10.3, 9.5, 8.6, 7.8, 7.5, 6.4, 6.4, 5.9]
weights=1.0/y ;Define a vector of weights
A=[10.0, -0.1, 2.0] ;Provide an initial guess of the function's
parameters
yfit=CURVEFIT(x,y,weights, A, FUNCTION_NAME='gfunct') ;Compute the
parameters
print, 'Function parameters: ', A
plot, x, y, yrange=[5,13], ystyle=1, xrange=[0, 11], xstyle=1, psym=1
ind=sort(x)
oplot, x[ind], yfit[ind], linestyle=1, color=5
END;PRO CURVE_FITTING
|
|
|
|
| Re: CURVEFIT for multiple datasets [message #58463 is a reply to message #58442] |
Thu, 31 January 2008 10:32  |
Spon
Messages: 178
Registered: September 2007
|
Senior Member |
|
|
On Jan 31, 4:13 pm, Paul van Delst <Paul.vanDe...@noaa.gov> wrote:
> chloesharro...@gmail.com wrote:
>> I'm an undergraduate trying to do an MPhys Project and
>> I'd rather try and code as much as possible myself so I can get credit
>> for it.
>
> Strange approach. Your ultimate goal should be to get the correct answer. That's much
> easier to do using software you can trust.
>
> Besides, regardless of whether you use IDL's CURVEFIT, or Criag's MPFIT, you're still
> using "advanced fitting programmes" that you can't take credit for. You can take credit
> for writing the code that feeds those procedures the data, I guess.
I completely agree - surely whether you use mpfitfun as programmed by
Craig or curvefit as programmed by the guys at RSI (as it probably was
when it was coded) makes no difference to how much credit you can take
for it? The only difference is that the RSI guys probably got paid for
their work!
Plus, as David mentioned, it's good for the soul, or something...
>> Luckily, I have found a way to do a fit for 2datasets now - by
>> puttting them all into one big array, so hopefully I can extend this
>> further so I put 2000+ datasets in one array and then fit an
>> exponential to that new array.
Could you expand please? I'm intrigued.
I only got as far as realising the elliptical input parameters to
GAUSS2DFIT were not what I was looking for before stumbling upon mpfit
which works very well for 2-dimensional datasets so long as you're
*really sure* you know what you're asking it to do. Never looked
back ;-)
>
> What about taking the average of all your datasets (assume the same abscissa values) and
> fit that using the std dev's of each point as an error estimate?
>
> cheers,
>
> paulv
take care,
Chris
|
|
|
|
| Re: CURVEFIT for multiple datasets [message #58471 is a reply to message #58463] |
Thu, 31 January 2008 08:13 |
Paul Van Delst[1]
Messages: 1157
Registered: April 2002
|
Senior Member |
|
|
chloesharrocks@gmail.com wrote:
> Hi Chris
>
> Thanks for your reply. I'm reluctant to use advanced fitting
> programmes as I'm an undergraduate trying to do an MPhys Project and
> I'd rather try and code as much as possible myself so I can get credit
> for it.
Strange approach. Your ultimate goal should be to get the correct answer. That's much
easier to do using software you can trust.
Besides, regardless of whether you use IDL's CURVEFIT, or Criag's MPFIT, you're still
using "advanced fitting programmes" that you can't take credit for. You can take credit
for writing the code that feeds those procedures the data, I guess.
> Luckily, I have found a way to do a fit for 2datasets now - by
> puttting them all into one big array, so hopefully I can extend this
> further so I put 2000+ datasets in one array and then fit an
> exponential to that new array.
What about taking the average of all your datasets (assume the same abscissa values) and
fit that using the std dev's of each point as an error estimate?
cheers,
paulv
|
|
|
|
| Re: CURVEFIT for multiple datasets [message #58474 is a reply to message #58471] |
Thu, 31 January 2008 06:37 |
David Fanning
Messages: 11724
Registered: August 2001
|
Senior Member |
|
|
chloesharrocks@gmail.com writes:
> Thanks for your reply. I'm reluctant to use advanced fitting
> programmes as I'm an undergraduate trying to do an MPhys Project and
> I'd rather try and code as much as possible myself so I can get credit
> for it.
Oohh, totally the wrong attitude. You should try to steal the
best ideas you can, build on those, and become famous for giving
ideas away. Much better karma that way. :-)
Cheers,
David
--
David Fanning, Ph.D.
Fanning Software Consulting, Inc.
Coyote's Guide to IDL Programming: http://www.dfanning.com/
Sepore ma de ni thui. ("Perhaps thou speakest truth.")
|
|
|
|
| Re: CURVEFIT for multiple datasets [message #58475 is a reply to message #58474] |
Thu, 31 January 2008 06:28 |
chloesharrocks
Messages: 16
Registered: January 2008
|
Junior Member |
|
|
Hi Chris
Thanks for your reply. I'm reluctant to use advanced fitting
programmes as I'm an undergraduate trying to do an MPhys Project and
I'd rather try and code as much as possible myself so I can get credit
for it.
Luckily, I have found a way to do a fit for 2datasets now - by
puttting them all into one big array, so hopefully I can extend this
further so I put 2000+ datasets in one array and then fit an
exponential to that new array.
Thanks for all your help
Chloé
|
|
|
|
| Re: CURVEFIT for multiple datasets [message #58476 is a reply to message #58475] |
Thu, 31 January 2008 06:15 |
Spon
Messages: 178
Registered: September 2007
|
Senior Member |
|
|
On Jan 31, 12:55 pm, chloesharro...@gmail.com wrote:
> Dear all
>
> I'm trying to fit an exponential curve to multiple data sets which
> have been overplotted on a graph. To get a feel for CURVEFIT (as I'm
> a novice at IDL and CURVEFIT), I wrote the following 2 programs (based
> on the one given in IDL help), which produce a nice exponential curve
> when there's only ONE dataset:
>
> =====
> PRO gfunct, x, t_a, f_t, PDER
>
> t_bx = EXP(t_a[1]*x)
> f_t = t_a[0] * (t_bx) + t_a[2]
>
> ;If the procedure is called with four parameters, calculate the
> partial derivatives
> IF N_PARAMS() GE 4 THEN $
> pder = [[t_bx], [t_a[0]*x*t_bx], [replicate(1.0, N_ELEMENTS(x))]]
>
> END
> =====
> and
> =====
> PRO CURVE_FITTING
> x=FLOAT(indgen(10))
> y=[12.0, 11.0, 10.2, 9.4, 8.7, 8.1, 7.5, 6.9, 6.5, 6.1]
> weights=1.0/y ;Define a vector of weights
> A=[10.0, -0.1, 2.0] ;Provide an initial guess of the function's
> parameters
>
> yfit=CURVEFIT(x,y,weights, A, FUNCTION_NAME='gfunct') ;Compute the
> parameters
> print, 'Function parameters: ', A
>
> loadcolors
> pson, filename='Curve_fitting.ps'
> plot, x, y, yrange=[5,13], ystyle=1, xrange=[0, 11], xstyle=1, psym=1,
> color=0
> oplot, x, yfit, linestyle=1, color=5
> psoff
> END
> =====
>
> How can I extend this for the case when I have say 2data sets, ie:
> x=[[indgen(10)], [indgen(10)]]
> y=[[12.0, 11.0, 10.2, 9.4, 8.7, 8.1, 7.5, 6.9, 6.5, 6.1], [11.5, 10.8,
> 10.3, 9.5, 8.6, 7.8, 7.5, 6.4, 6.4, 5.9]
> and I want to plot an exponential curve through all of this data?
> Eventually, I would like my program to be able to plot an exponential
> curve through 2000+ data sets all of which have been plotted on the
> same graph and seem to show an exponential trend.
>
> Thanks very much
> Chloé
This doesn't really answer your question in any way, but if you're
going to do a lot of curvefitting, I'd suggest having a look at Craig
Markwardt's site:
http://cow.physics.wisc.edu/~craigm/idl/idl.html
His fitting programmes are faster and more robust than CURVEFIT in my
(admittedly limited) experience, and it's certainly true that they
don't crash and burn as easily as CURVEFIT does.
To try and answer your question, there's no way I can see of fitting
more than one datapoint at a time using the method you describe. The
brute force way involves wrapping the lot in (slow) FOR loops. The
elegant and efficient way, if there is one, eludes me.
There may well be a way to use Craig's mpfit to help you here, but I
don't know - perhaps someone else does?
> Eventually, I would like my program to be able to plot an exponential
> curve through 2000+ data sets all of which have been plotted on the
> same graph and seem to show an exponential trend.
It may also be worthwhile asking yourself if you really need a fitted
curve to every single data set; or one curve that best fits the whole
data (in which case I'd suggest looking at MOMENT or TOTAL, for
example, before you do any curve fitting at all.) Maybe even something
in between?
Hope this helps,
Chris
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
