On Thu, Jul 24, 2003 at 04:07:17PM -0700, Ian Clarke wrote:
> > Let me clarify:  The vertical axis is actual or predicted latency.
> > The horizontal is just "sample index".
> 
> Why are the first few BDA1 predictions along the X axis all the same in
> the first graph?

When presented with a new data point that goes into a bin with
no previous data, the BDA1 algorithm selects the closest bin which
contains data.  Here are the first five points from inp.dat:
0 115511 2395
2395 61646 10133
12528 26206 535
13063 48597 958
14021 115511 2203
Remember, BINDIVISOR is set to 10000 for this run.
The first point goes in bin 11 because 115511 / 10000 = 11.
Now the algorithm predicts the latency for the second point, and since
the only data it's seen so far is 2395 from point 0, it correctly
predicts the same.
The next point (the outlier) goes in bin 6 because 61646 / 10000 = 6.
So now there are two averages initialized: 
Bin 11  at 2395 and Bin 6 at 10133.
The third point is in bin 2 because 26206 / 10000 = 2.
Since there is no information for bin 2, it uses the closest bin that
has info, which is bin 6, which is the outlier, so it predicts 10133.
On the fourth point, it looks at bin 4.  Same problem, no info, so it
searches outward from bin 4 and finds bin 6 so predicts 10133 again (sample 3).
(it could have chosen bin 2 also but I check the + side first)
Next it looks at the fifth point which fits in the same bin
as the first point, so it outputs the average in bin 11, 2395.

Rudi


_______________________________________________
devl mailing list
[EMAIL PROTECTED]
http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl

Reply via email to