PDA

View Full Version : Re: Woltring's quintic spline



Jesus Dapena
10-27-1996, 11:52 PM
Dear Biomch-L readers:

Ton van den Bogert and Juan Vicente Dura have pointed out that I
should use Mode 3 (and NOT Mode 1) in Woltring's programs if I am providing
the desired average squared deviation in VAL. I have now done this, but
there still is a problem:

Running my data with Mode=3 and with VAL=0.000010, the sum of
squares came out to be ssq = .000066 which is now small, much smaller than
the goal of: (42*0.000010=) 0.000420. So now the program is being far too
conservative, and does not smooth as much as I wanted. It is not that the
data are already so smooth (with the ssq of 0.000066) that no further
smoothing is possible under quintic spline. I think this is clear from the
results that I obtained when I tried a much larger value of VAL (0.001).
Then, I got ssq = .006823, which is a lot larger than when I tried
VAL=0.000010. This seems to indicate that the smoothing was far from
"saturated" when I used VAL=0.000010, and therefore with that value of VAL
the program should have smoothed more, getting nearer to the requested ssq
value of 0.000420.

In sum, the problem now is the reverse from when I posted my
previous message: The Woltring subroutines are not smoothing the data
enough; the sum of squares of the residuals is way below what I requested in
my input parameter (VAL). While technically this does not go against the
requirement that the sum of squares of the deviations be smaller than N*VAL,
it goes too far in the opposite direction, leaving the data too unsmooth.

Is this a standard occurrence when using Mode 3 with Woltring's
programs?


(For a long time, I have used a quintic spline program devised by
Les Jennings, and with these same data and the requested average squared
deviation set at 0.000010 the Jennings program smooths the data until a sum
of squares of 0.000420 is reached. The problem with the Jennings program is
that in the computer that I regularly use, sometimes it does not return ANY
values! So for most data sets the Jennings program is fine, but with some
it gives me trouble. That is the reason why I am checking out Woltring's
quintic spline, but up to here I am not happy with the results.)

Jesus
---
Jesus Dapena
Department of Kinesiology
Indiana University

Bloomington, IN 47405, USA
1-812-855-8407
dapena@valeri.hper.indiana.edu
http://ezinfo.ucs.indiana.edu/~dapena