[mnet-devel] analytical solution to the FEC reliability problem

[Zooko]

> For 32, 16, and 0.5 we get:
> 
> choose(M, K) = 601080390
> P(S)^K * (1-P(S))^(K-M) = 1.0   (hrmmm....)

Yeah, this is the kind of formula that gets eaten up by imprecision.  We're 
multiplying a very large number times a very small number times a very small 
number in order to get a number between 0 and 1.  If the small numbers drop 
precision, the answer can come out quite wrong.

(Note: it also might be that my formula is wrong, but the fact that you get 1.0 
when we expect 0.99something suggests that precision is a problem.)

Possible solution: change simul.py to use fixedpoint and to test harder for 
convergence before returning.  The "simulator" ("Monte Carlo") method doesn't 
suffer from imprecision as much as the analytical one does.

--Z



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
mnet-devel mailing list
mnet-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mnet-devel