Re: Non-linear Rgression
- To: mathgroup at smc.vnet.net
- Subject: [mg71637] Re: Non-linear Rgression
- From: Bill Rowe <readnewsciv at sbcglobal.net>
- Date: Sat, 25 Nov 2006 05:37:18 -0500 (EST)
On 11/24/06 at 1:17 AM, amin67r at gmail.com (aminr) wrote:
>Hello, I'm a physics student.I have done an experiment .I want to
>use Mathematica , I found a problem in using it :
>for example I have data ={ {0,0},{1,2},{3,6}} and I want to fit
>them for a equation such as " y=Log[Cosh[a*x]] " so I want the value
>of parametre"a" and also its error(i.e. a=0.045+- 0.05 then error
>is 0.05 )
>and also the regression (i.e. R=0.995 )
>please ,please , tell me how I can do it ?
If only the best fit parameters are needed, then the built in
function FindFit does what you need.
For example;
In[19]:=
data={{0,0},{1,2},{3,6}};
In[20]:=
FindFit[data,Log[Cosh[a x]],{a},x]
Out[20]=
{a -> 2.2753570110148678}
To easily get confidence intervals etc, the best choice would be
to use the non-linear regression package that is part of the
standard add ons. That is:
In[21]:=
<<Statistics`
In[22]:=
NonlinearRegress[data, Log[Cosh[a*x]], x, {a},
RegressionReport -> {BestFitParameters,
ParameterCITable, AsymptoticCorrelationMatrix}]
Out[22]=
{BestFitParameters -> {a -> 2.2753570110148678},
ParameterCITable -> TableForm[
{{"", "Estimate", "Asymptotic SE", "CI"},
{a, 2.2753570110148678, 0.09599842662397484,
{1.8623091186495693, 2.6884049033801674}}},
TableDepth -> 2, TableHeadings ->
{{a}, {"Estimate", "Asymptotic SE", "CI"}}],
AsymptoticCorrelationMatrix -> MatrixForm[{{1.}}]}
Note a variety of fit statistics are available when using
NonlinearRegress. See the documentation for details.
--
To reply via email subtract one hundred and four