Smoothing

[Roedy Green]

I wondering if there is a better way to handle smoothing.

Examples of the raw data would be how many hits you had each day on a
web page or how much you weighed on a given day or what your % bodyfat
was on a given day.


What I have been using is a moving average, 3 days before and 3 days
after This works reasonably well, except at the end where you have no
future data past today to average in. I wanted to factor out the
weekly pattern of hits (low on the weekend).

Another problem I have is not taking the hit measurement at exactly
the same time each day. I thought it might be possible to record the
time the hit was taken, and somehow adjust it based on the past
history of how hits typically arrive at different times of the day.

--
Roedy Green Canadian Mind Products
http://mindprod.com
PM Steven Harper is fixated on the costs of implementing Kyoto, estimated as high as 1% of GDP.
However, he refuses to consider the costs of not implementing Kyoto which the
famous economist Nicholas Stern estimated at 5 to 20% of GDP

 

[Eric Sosman]

I wondering if there is a better way to handle smoothing.

Examples of the raw data would be how many hits you had each day on a
web page or how much you weighed on a given day or what your % bodyfat
was on a given day.

What I have been using is a moving average, 3 days before and 3 days
after This works reasonably well, except at the end where you have no
future data past today to average in. I wanted to factor out the
weekly pattern of hits (low on the weekend).

Another problem I have is not taking the hit measurement at exactly
the same time each day. I thought it might be possible to record the
time the hit was taken, and somehow adjust it based on the past
history of how hits typically arrive at different times of the day.

The general topic area is "filtering," and I'd suggest
trying the question in a mathematics or signal-processing
forum. Some image-processing procedures may also be of
interest, although they tend to filter in two dimensions
as opposed to your one-dimensional examples.

 

[Lew]

The general topic area is "filtering," and I'd suggest
trying the question in a mathematics or signal-processing
forum. Some image-processing procedures may also be of
interest, although they tend to filter in two dimensions
as opposed to your one-dimensional examples.

Also research "proportional-integral-derivative" (PID) filtering, a technique
that captures trends without being too sensitive to short-term variations or
too insensitive to actual trends.