[Numpy-svn] r3330 - trunk/numpy/lib
numpy-svn at scipy.org
numpy-svn at scipy.org
Fri Oct 13 14:37:50 CDT 2006
Author: charris
Date: 2006-10-13 14:37:48 -0500 (Fri, 13 Oct 2006)
New Revision: 3330
Modified:
trunk/numpy/lib/polynomial.py
Log:
Mention scaling in the polyfit docstring.
Modified: trunk/numpy/lib/polynomial.py
===================================================================
--- trunk/numpy/lib/polynomial.py 2006-10-13 19:26:43 UTC (rev 3329)
+++ trunk/numpy/lib/polynomial.py 2006-10-13 19:37:48 UTC (rev 3330)
@@ -201,16 +201,21 @@
value method takes a paramenter, 'rcond', which sets a limit on the
relative size of the smallest singular value to be used in solving the
equation. The default value of rcond is the double precision machine
- precision as the actual solution is carried out in double precision.
+ precision as the actual solution is carried out in double precision. If you
+ are simply interested in a polynomial line drawn through the data points
+ and *not* in a true power series expansion about zero, then the best bet is
+ to scale the x sample points to the interval [0,1] as the problem will
+ probably be much better posed.
WARNING: Power series fits are full of pitfalls for the unwary once the
- degree of the fit get up around 4 or 5. Computation of the polynomial
- values are sensitive to coefficient errors and the Vandermonde matrix is
- ill conditioned. The knobs available to tune the fit are degree and rcond.
- The rcond knob is a bit flaky and it can be useful to use values of rcond
- less than the machine precision, 1e-24 for instance, but the quality of the
- resulting fit needs to be checked against the data. The quality of
- polynomial fits *can not* be taken for granted.
+ degree of the fit get up around 4 or 5 and the interval of sample points
+ gets large. Computation of the polynomial values are sensitive to
+ coefficient errors and the Vandermonde matrix is ill conditioned. The knobs
+ available to tune the fit are degree and rcond. The rcond knob is a bit
+ flaky and it can be useful to use values of rcond less than the machine
+ precision, 1e-24 for instance, but the quality of the resulting fit needs
+ to be checked against the data. The quality of polynomial fits *can not* be
+ taken for granted.
For more info, see
http://mathworld.wolfram.com/LeastSquaresFittingPolynomial.html,
More information about the Numpy-svn
mailing list