Can I use "fitnlm" for multiple regression ? and How?

Hi everybody,

I have mutiple nonlinear regression like (x1,x2,x3 and y) the equation is "y=x1+x2^n+x3^n" , where n is known, , can I use "fitnlm" function and how can set up it?

Thyanks in advance,

Riyadh

[dpb]

Is n integer or fractional?

[Riyadh Muttaleb]

n is fractional.

[dpb]

Then, no, fitlm can't handle it.

You have two choices --

1. if have the Curve Fitting TB, you can specify a custom linear model with fittype
2. you can write the design matrix specifically and use backslash operator to solve for coefficients

[Riyadh Muttaleb]

Thank you!

[Image Analyst]

If the x are your measurements (training data) and n is some known fractional value, then what are your unknown parameters?

Is there a weighting factor in front of each x that multiplies it??? Likd "y = c1 * x1 + c2 * x2^n + c3 * x3^n" where the c are the unknowns to be solved for????

Or did you mean that actually n is unknown (to be solved for) and there are no weighting/scaling factors multiplying the x?

By the way, in cases like this I'd use fitnlm(). In fact I always use fitnlm() because I think it's more general and can automatically handle the linear case when needed.

[Riyadh Muttaleb]

I wrote the equation that I want to solve, the regeression gives the constants (a,b,c, and d)

y=a+b*x1+c*x2^n+d*x3^n;

IF this equation can be solved with "fitnlm" PLEASE let me know how? n is fractional (from 0.1-3)

Answer 1

[dpb]

Well, I take that back...actually you could do it with fitlm if you do a change of variable...

u1=x2.^n; 
u2=x3.^n;
t=table(x1,u1,u2,y);
lm=fitlm(t,'y~x1+u1+u2');

I've used fitlm very little, so I'm sorta' shootin' in the dark, but I believe that's correct.

I can't figure out the doc to use the variables outside the table; it's the most confusing description and lacking in useful examples... :(

[Riyadh Muttaleb]

Thank you so much, actually I have used "fitlm",

[dpb]

Well, post back if this works...and if you know how to use just the defined variables, I'd be interested in seeing that syntax of how to reference the X,Y arrays in the model spec...

It's just too confusing for occasional use; I suppose if one works with it continusously one would learn the nuances and how the designers intended it to work.

[dpb]

[OP Answer moved to comment...dpb]

Sure,

lm=fitlm([x1 x2 x3],y) 
where x2=x^n, x3=u^n