polyfit and polyval problem

Hi there,

I am trying to plot a line using the ployfit and command

x and y are the coordinates in attached xy the plot.

I can't plot a polyfit curve that reproduces the original.

I alway came up with something different from the original curve.

[Star Strider]

Perhaps if you post your code, we can help you find and solve the problem.

[dpb]

Well, there's absolutely no chance the figure was a polynomial that generated it....where did you get it from???

Is there any physical basis for the data that would give a klew as to what the underlying model should be?

[Paul Rogers]

p1=polyfit(x,y,3);
x_intrest=[0:0.01:0.8];
y_intrest=polyval(p1,x_intrest);
plot(x_intrest,y_intrest)

[Paul Rogers]

yes I got the y from a more complex equation, datas in y are the evalueted point I then wanted to approximated

[Guillaume]

Polynomial fitting is not magical. If the function you want to fit is nothing like a polynomial, as you have here, there's no way you'll find a good fit.

You'll have to try another kind of fit as dpb said and to know what that fit is you need to know some details about what your data represent.

[Image Analyst]

Can you attach a screenshot? I can't open the .fig file directly from Firefox - it throws an error message when trying to launch MATLAB to display the figure even though MATLAB is already running.

[dpb]

Here ya' go, IA...

[dpb]

"I got the y from a more complex equation, ..."

Well, there's the place to start. What makes sense to approximate whatever that is/was.

What restrictions are on the approximation used, if any?

Answer 1

[Paul Rogers]

here we are, I think I found the solution.

I can select only a portion of that curve and take the x and y values,

then the code it's this

p1=polyfit(x,y,4);
t_intrest=[0.2:0.1:0.4];
y_intrested=polyval(p1,t_intrest);
plot(x,y);
hold on
plot(t_intrest,y_intrested);
hold on

In attaced the new x and y.

Now it's up to me to find the largest portion possible where I can use polyfit.

[dpb]

"Now it's up to me to find the largest portion possible where I can use polyfit."

Why on earth is that so????

That section would fit a quadratic almost perfectly but still why not use a form that fits the data instead instead of trying to make it complicated?

>> N1=201;N2=401;
>> b=polyfit(x(N1:N2),y(N1:N2),2);
>> hL=plot(x(N1:N2),polyval(b,x(N1:N2)),'r-');
>> xlim([0.15 0.45])
>> 

[Paul Rogers]

the thing is that I am trying to use the polyfunction for the larger portion possible of this curve

[dpb]

That's just not feasible.

[Guillaume]

I'm not sure how many times we can say it, your curve is nothing like a polynomial and attempting to fit a polynomial to it is the wrong approach.

To me it looks like your curve may be a log-normal curve. Again, some information on what created that curve could tell you that.

[Paul Rogers]

I am trying to find a polynomial approx for that curve. I don't mean on its ooverall domain, it would be good even on a portion of that.

For example, I found that from 0 to 0.8 (more or less), it could be use this method.

P.S.

Thanks everyone for the help.