[Date Index]
[Thread Index]
[Author Index]
Re: Re: regress versus fit - force throughzero/forceconstant term to zero
*To*: mathgroup at smc.vnet.net
*Subject*: [mg61613] Re: [mg61574] Re: [mg61493] regress versus fit - force throughzero/forceconstant term to zero
*From*: Chris Chiasson <chris.chiasson at gmail.com>
*Date*: Sun, 23 Oct 2005 05:46:44 -0400 (EDT)
*References*: <acbec1a40510192358g44579892k5d78ab545b45e151@mail.gmail.com> <002901c5d556$c37f97f0$0401a8c0@achirana> <200510220911.FAA14997@smc.vnet.net>
*Sender*: owner-wri-mathgroup at wolfram.com
Thinking about this more:
Does it really make sense to have significance/precision associated
with the best fit parameters?
What is the best way to take into account the variance in the best fit
parameters due to the fact that the model might be not 100% correct
**and** the fact that each data point is probably not infinitely
precise?
On 10/22/05, Chris Chiasson <chris.chiasson at gmail.com> wrote:
> It seems like the diagnostics returned by Regress are okay, but they
> don't have any significance arithmetic ability. Looks like I will be
> doing each problem twice... once for the regression report, and once
> for the "significance arithmetic" available with Fit.
> David, do you work for WRI? You said you looked at the "code" for
> Regress... I was thinking about filing a bug report or RFE to have
> Regress' BestFit return the same result as Fit for "bignum"/"sig.
> arithmetic" inputs. Good idea y/n?
>
> On 10/20/05, David Annetts <davidannetts at aapt.net.au> wrote:
> > Hi Chris,
> >
> > > Thank you Dave, that was most timely and informative.
> > > Your reward is:
> > >
> > > another question :-[
> > >
> > > << Statistics`LinearRegression`
> > > FAASCalPts = {{1, 0.032`2}, {2, 0.067`2}, {5, 0.183`3}}
> > > FAASCalFit = Fit[FAASCalPts, {ñCaRatio}, ñCaRatio]
> > > FAASCalRegression = Regress[FAASCalPts, {ñCaRatio}, ñCaRatio,
> > > \ RegressionReport -> {SummaryReport, BestFit},
> > > IncludeConstant -> False] FAASCalRegression =
> > > Regress[FAASCalPts, {1, ñCaRatio}, ñCaRatio, \
> > > RegressionReport -> {SummaryReport, BestFit}]
> > >
> > > The fit points now have their precision specified (well, some
> > > of them do, the integers are assumed to be at infinite precision).
> > >
> > > Do you have any thoughts about why Fit seems to give a result
> > > that keeps precision in mind (though it seems to me like
> > > there should be more digits than just 0.4), but Regress is
> > > giving out answers as if I had fed it high precision numbers?
> >
> > ... Just had a look at the code for Regress. It seems (from the code in
> > vRegress) that precision in Regress is taken from the precision of the
> > basis.
> >
> > You might try using a DesignMatrix to a similar behaviour to Fit, for
> > example,
> >
> > DesignMatrix[FAASCalPts, {0, ñCaRatio}, ñCaRatio] (* change 0->1 to
> > include the constant *)
> > DesignedRegress[%, Last@Transpose[FAASCalPts]]
> >
> > But it seems that you'll have to work a bit to extract something as basic as
> > BestFit.
> >
> > Regards,
> >
> > Dave.
> >
>
>
> --
> Chris Chiasson
> http://chrischiasson.com/contact/chris_chiasson
>
>
--
http://chrischiasson.com/contact/chris_chiasson
Prev by Date:
**Re: Re: How smooth graphs?**
Next by Date:
** Re: aggregation of related elements in a list**
Previous by thread:
**Re: regress versus fit - force throughzero/forceconstant term to zero**
Next by thread:
**Re: Re: Re: regress versus fit - force throughzero/forceconstant term to zero**
| |