On 11/5/10 1:07 AM, Alessandro Vesely wrote:
On 04/Nov/10 20:49, Darxus(_at_)chaosreigns(_dot_)com wrote:
> On 11/04, Alessandro Vesely wrote:
>> On 02/Nov/10 21:39, Darxus(_at_)ChaosReigns(_dot_)com wrote:
>>> Now I'm thinking about providing non-spam percentages, and a
>>> number indicating relative total volume of email from the IP,
>>> probably both with a range of 256 values. Basically all the
>>> precision I think anybody could find useful.
>> The second format might allow some sort of projective
>> conglomeration of results.
> What?
If I have reports from both A and B about S, I may want to add their
results. I can extrapolate the real volumes Va and Vb that A and B
got from S by scaling the relative volumes Ra and Rb by, say, the
volumes I send to A and B, respectively. Call Ma and Mb these
"projections" of A and B on My site, and Pa and Pb the percentages
found in the report. Then I can reckon their sum as
(Pa*Va + Pb*Vb) / (Va + Vb) = (Pa*Ra*Ma + Pb*Rb*Mb) / (Ra*Ma +
Rb*Mb)
One may want to season the projection factors with more ingredients,
such as the trust granted to the relevant reporter.
Not all email is bidirectional, where an amount sent will not reflect an
amount received. Received rates are poor predictors of sending rates
due to nonlinear relationships between sending and receiving. People
are able to read only so much. Volume over time can be expressed as
absolutes using natural log with as little as 4 bits. As sending volume
increases, a receiver's relative volume will decrease due to a required
greater diversity of sending destinations. Natural log can approximate
receiver reductions resulting from a greater diversity of destinations.
Summing log volume requires conversion back to linear values adjusted to
the same time interval. When not converted, natural log might better
predict expected individual receive volumes when a very large number of
receivers have been aggregated to better approximate actual sending rates.
-Doug
_______________________________________________
Asrg mailing list
Asrg(_at_)irtf(_dot_)org
http://www.irtf.org/mailman/listinfo/asrg