cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
Showing results for 
Search instead for 
Did you mean: 

Power and Exp Regression?

Power and Exp Regression?

Why are guess values needed in MathCAD for power and exponential regressions? Excel doesn't need them. Attached is a mathcad and excel document that gets me nothing but frustrated. All I have is x and y values that I am passing into both MathCAD and excel. MathCAD can't find the a,b and c values to satisfy the regressions, even when I give it the same ones I found using Excel. I even graphed the excel values in Mathcad and they worked. Why doesn't Mathcad find the coefficient values of the regression?

Matt Pittard
16 REPLIES 16

Power and Exp Regression?

1. The function expfit, pwrfit etc solve *nonlinear* task and cann return two or more answer. Thay need guess values.
2. Statistic in Excel is very bad.

Valery Ochkov
http://twt.mpei.ac.ru/ochkov

Power and Exp Regression?

Yeah, Excel has terrible Statisticial regression analysis. They don't print out nearly enough significant figs. My problem was that MathCAD wasn't finding one of the nonlinear solutions.

Matt

Power and Exp Regression?

Origin is superb on some aspects.
It was no sweat fitting your data.
By the way, some of the data are in error. In such case, it's good practice to eliminate the bad ones. If your project was " real life " then use it. I think this the true approximation .

Power and Exp Regression?

Origin is superb on some aspects.
It was no sweat fitting your data.
By the way, some of the data are in error. In such case, it's good practice to eliminate the bad ones. If your project was " real life " then use it. I think this is the true approximation .

Power and Exp Regression?


Try these values

Chi2 R2
---------------------
1.43763 0.99916

A1 = 2139.745
t1 = 1.2507
A2 = 171.36
t2 = 3.832

Regression - visual control of guess values

Hi! See:

PS I think that Excel has guess values too - zeros.
Valery Ochkov
http://twt.mpei.ac.ru/ochkov

Regression - visual control of guess values

The approximation I gave yesterday, was extracted using Origin. This software does not require initial guess values. By default it varies each parameter. But the number of iterates may be specified. The first approximation was with 1 iterate, the second one used 10 iterates.

Regression - visual control of guess values

Hi,

sorry to disagree about Origin, it too uses guess-values!

It obviously uses good default-guess-values, that you don�t see, when just fitting a curve through data using the menu for the simple fitting.

But if you go into NLSF (non-linear fitting) and check "more/advanced", you will see the according equations, their parameters and start-values! (Bad guess, bad fit, that easy).

Origin is really great for displaying and analyzing data!

Hubert

Regression - visual control of guess values

You are right Hubert. Since I have Mathcad, I rarely use Origin. You like it too ! So nice graphs. Combining Origin and Mathcad, would be near perfect.
I tried Origin for one of my favorite rational fraction approximation, then it has no guess values because there is no model. It comes close to exact answer, just by itself, and there are 3 coefficients. I do the same with Mathcad, but in steps. It could be programmed in Mathcad, but I'm not skilled at that.
As a matter of facts, for certain type of curves, I can generate an excellent P(x)/Q(x), in simple steps.
Why these big guns: Mathcad, Mathematica, Mathlab... are so poor about approximating functions ???
Can you imagine, none of them included the Thi�l� Continued fraction...F.
If you read me, Danish Collab, I will make a file for you. Because Thi�l� is Danish. And such a great tool, you must have it.
I have many sheets, using Thi�l�. Yes Sir !

Regression - visual control of guess values

Guys--where do I get a copy of this "ORIGIN 6.0"--how much does it cost?

Matt

Regression - visual control of guess values

I copied the page that remained in my mail box.
The latest version is 6.1.
In 6.1, you will find added functionalities. As such, it's the same software.
Their forum is well, well, well supported.
Superb companionship with Excel. Origin is a spread sheet type too.
But not very convenient writing equations in script format.
I have not tried it with Mathematica.
Origin is a very interesting gear.
I believe < $ 1000.
Your link is at the bottom of the page.
Bob Sullivan will gladly consider your quest.
I think their latest link is:
http://www.origin.lab

Let me know if you have difficulties.










ORIGIN LINK FOR MATHEMATICA REVIEW ON SCI WEBSITE
ORIGIN 6 RATED THE PERFORMANCE LEADER COMPARED TO SIGMAPLOT 5 AND AXUM 6
FREE TOOLS TO EXPAND ORIGIN'S FUNCTIONALITY NOW AVAILABLE
REVIEW OF ORIGIN 6 BY PHYSLINK - 5 STAR RATING!
ORIGIN CONTEST - WIN A FREE COPY OF ORIGIN!


*****************
SCI'S EDITOR'S CHOICE: ORIGIN LINK FOR MATHEMATICA
Scientific Computing and Instrumentation Magazine recently highlighted the Origin Link for Mathematica on their website. This latest tool for Origin and Mathematica users includes an intuitive and versatile multi-tabbed dialog box through which data and expressions can be transferred between Mathematica and Origin. Use any of Mathematica's functions to analyze your data, then use Origin to graph the results. This Link is a free download - but you must have Origin 6 or the Origin 6 demo and Mathematica 4. Try it out today!
http://www.scimag.com/scripts/RelSection.asp?RPTID=RELSECTION&RELTYPE=PP
http://www.microcal.com/www/applications/olm.asp

*****************
ORIGIN 6 RATED THE PERFORMANCE LEADER IN DATA ANALYSIS AND TECHNICAL GRAPHICS SOFTWARE CATEGORY
The National Software Testing Laboratories (NSTL) tested Origin 6 against SPSS SigmaPlot 5 and MathSoft Axum 6 and found Origin 6 to be the fastest in all aspects tested. The test compared the performance of the three softwares in five key aspects: Importing, Exporting, Plotting, Plot Formatting, and Data Analysis. For the full report, check out the link below:
http://www.microcal.com/www/press/pressnstl6.asp

******************
FREE CUSTOM TOOLS AVAILABLE TO EXPAND ORIGIN'S FUNCTIONALITY
Check out these new tools available as free downloads that have either been created by our users or by the technical staff at Microcal Software. Downloads include a tool to import and analyze nuclear magnetic resonance (NMR) data; a specialized tool to read, extract and analyze data stored in spec files, a popular file format for data collection from X-ray and neutron scattering experiments; as well as a script tool for programming LabTalk which includes 20 well documented script examples as well as a convenient interface for organizing, selecting, displaying, and executing scripts. Check back regularly for additional free downloadable tools for Origin, or submit your own. Click here for more information:
http://www.microcal.com/www/applications/index.asp

*****************
ORIGIN 6 GETS 5 STARS - THE HIGHEST RATING - FROM PHYSLINK!
Physlink, the physics and engineering online reference website, rated Origin 6 as "EXCELLENT" in its latest review. Quote: "two most important qualities that make Origin distinct from the other data analysis and presentation software are it's beautifully intuitive interface and it's mighty versatility." A direct link to the review is below:
http://www.microcal.com/www/origin60/reviewplink.asp

******************
WIN A FREE COPY OF ORIGIN!!
Submit your Origin graphs from now until March 31st for our Graph of the Month contest, and you may be chosen to win a free copy of Origin. You may submit graphs created with the Origin demo. For more information, click below:
http://www.microcal.com/www/marketing/index.asp

******************
Thank you for taking the time to evaluate Origin - the fastest software for scientific graphing and analysis. For more information on pricing or on the specific features of Origin, please contact our Technical Sales Department at sales@microcal.com, or contact your local dealer. If you do not want to recieve future emails on Origin, please reply to this email with "Remove from Mailing List" in the subject line.

Microcal Software
http://www.microcal.com

Regression - visual control of guess values

Sorry. But it is the chat about Mathcad. I know a lot of software for regression analyze (best than Excel, Origin etc). I can tell about this software, but thru e-mail.
Valery Ochkov
http://twt.mpei.ac.ru/ochkov

Regression - visual control of guess values

Hi,

well, I second Valery, about this forum rather belonging to Mathcad, regarding the Origin-ads. Yet a discussion about other programs and comparing their capabilities and limitations to Mathcad is useful too! 😉

One final info about Origin:
it�s great for displaying and analyzing data. The manufacturer is OriginLab (formerly known as Microcal), Origin6.1 is their most recent version (I always would recommend the PRO) and their website is www.originlab.com . They provide Demo-versions.

It would be great, to have some examples, which combine Mathcad with Origin (should be possible with scripting, but I am no OLE- or VB-wizard).

Cheers,
Hubert

Regression - visual control of guess values

Thanks Valery and Hubert.
I got used to approximate functions through my profession in Instrumentation. My tools are quite simple: Polynomials, Thi�l� (which I love so much and complements so nicely Lagrange).
I have a short library of remarkable functions.
In complement I wish I would have:

1. a tool to extract the 'an' coefficients for ChebyShev ( input [x : y ] and return the final an ....something like the approximation of Neperian Log. We have a case in cryogenic semiconductors)
2. For certain type of curve fitting, I have my own unpublished tool to extract rational fractions:
P(x)/Q(x). But I don't have a generalized method.

3. The issue is to have Mathcad execute in simple modules. My Fortran is so far away.

All suggestions are welcome.
Thanks: Jean

Regression - visual control of guess values

Sorry! But I does not say that the *visible* guess values are the best guess values. I only say that we can use the visualization for solve this task.
See my messages about this subject in Probability & Statistic Forum too.
Valery Ochkov
http://twt.mpei.ac.ru/ochkov

Power and Exp Regression?

All these people looking at this and coming up with fancy theories as to the problem. An no one bothering to look at the actual code and notice the the on error construction is coded backwards, so that the fitting command would only be executed if the evaluation of the guess values fail (not likely, as the guess values are coded as constants.

Tom Gutman
Announcements
Check out the latest
Mathcad Tip
"PTC Mathcad 15 / Prime 1-6 Update."