[SciPy-User] How to estimate error in polynomial coefficients from scipy.polyfit?
Mon Mar 29 12:47:15 CDT 2010
On Mon, Mar 29, 2010 at 1:31 PM, David Goldsmith
> On Mon, Mar 29, 2010 at 7:34 AM, Robert Kern <email@example.com> wrote:
>> On Sun, Mar 28, 2010 at 22:26, David Goldsmith <firstname.lastname@example.org>
>> > On Sun, Mar 28, 2010 at 11:36 AM, <email@example.com> wrote:
>> >> Crack permeability goes like the third power of the opening (that is,
>> >> fluid flow through cracks - think gas or oil in a fractured rock).
>> > Power law or polynomial: from a regression stand point, there's quite a
>> > big
>> > difference.
>> "Third power" == "x**3". He's not talking about a power law.
> Yes, I know that, but from a regression stand point, unless there's an
> offset (constant) term (in which case a two parameter polynomial fit is what
> you'll be doing) if your model is simply y = ax**3, aren't you better off
> doing the regression as if you were doing a power law, albeit w/ a fixed
> power (i.e., log transforming the data first, fixing the slope parameter at
> three, and then regressing to find the constant term, i.e., log(a))?
> In other words, I was soliciting examples of situations where a true
> polynomial (as opposed to a monomial) model was appropriate - I maintain
> that if your model is a monomial (integer power law model, w/ only one
> term), then, as far as regression is concerned, it is more appropriate to
> think of it as a power law model w/ a fixed parameter, not as a polynomial
> model. From this perspective, the number of "naturally occurring"
> polynomial models is greatly reduced.
That looks to me like splitting hairs.
It depends on the statistical model for the regression error, e.g.
y = ax**3 + u where u is normal, additive noise,
y = ax**3 *z where z is log-normal, multiplicative noise
ln(y) = ln(a) + 3*ln(x) + u with u = ln(z)
I would do it if I want to estimate or test if 3 is the correct power,
but not if 3 is known.
>> Robert Kern
>> "I have come to believe that the whole world is an enigma, a harmless
>> enigma that is made terrible by our own mad attempt to interpret it as
>> though it had an underlying truth."
>> -- Umberto Eco
>> SciPy-User mailing list
> SciPy-User mailing list
More information about the SciPy-User