If you like one of the applications you see, you can download a working
demo or free software from off the page by clicking on the disk beside
the file.
Do you know of a book I get hold of that would have some information on
'?fuzzy weighting?' (is there such a thing?) a neural net?
I now use NeuroWindows, by Ward Systems, with VB. It is an excellent
package, and I highly recommend it. If anyone has a project for which you
wish to use a Kohonen back-prop, I have developed some permutation
algorithms for self sensitizing the net. Since I believe net design to be
somewhat of an art form, this is my answer to a hesitant intuition. I can
not vouch for this approach based on an actual system I have developed
for anybody. However, I've used it to play with chaotic systems and got
some very interesting results.
For the fun of it, I started playing with some lottery data I had
analyzed while taking a statistics course at the U. of C. I noticed that
predictions for a 6/49 lottery system I sensitized, using this method,
were consistently within a very low standard deviation from the predicted
norms. I nearly drove a friend to madness when I gave him numbers for two
draws and all the numbers were within one from the actuals.
Granted, there is an amazing amount of coincidence here, but winning the
lotto number needs even more. Since the data is reordered, this defies
conventional methods of predicting probabilities of a randomly generated
sequence of numbers and makes a poor model to work with, so I thought it
best to design something that could make a use of this data rather than
try to feed it more analytical inputs.
Using combination routines, I had the net predict larger combinations
that included the highest number of hits for actuals that lay within the
deviational range of the predictions. I found that you would need at
least an 18 number combination of numbers to 'gaurantee' a statistical
percentage of the wins. That would also be the limit to what you could
afford to play. It would cost nearly $20K per roll to play the system. I
predicted 18 numbers in Jan. 92 and then tracked them. I quit tracking
them in Oct. of last year, as they appear to have been realized, but they
made some good guesses.
To play this method, would have cost $2m. per year. In the first year,
the recorded winnings exceeded $25m. And me too broke to play. Damn. If
you want to check the numbers out, check them against the last couple of
years winnings for the Canadian 6/49 starting in 01/01/92. If you're
interested, check out what they do, before that date. Interesting. In
order of weight, with (weight), the numbers are:
21(5),16(4),17(4),18(3),25(3),26(3),27(3),29(3),34(3),20(2A),7(2),14(2),2
8(2),31(2),36(2),37(2),39(2),12(1)
For reasons of which I am aware, these numbers have lost any predictive
validity they ever did possess. They need to be rerolled, but I am not
sure what effect time has had on the model. In a sense chaos must have
increased from inflationary expansion of probabilities within the model.
I am not sure that I have referred to the elements I am describing using
proper terminology, as I am not much of a mathematician, but in my mind,
I am charting multiple, predicted 'courses' through a hyper-expanding
model of probabilities.
If you think 18 numbers gives you a pretty good shot, go ahead and pick
any 18 at random and then check them against winning numbers. In fact,
there are more combinations of 18 than there are of 6. There are nearly
14m comb. of 6 and nearly 1.2q! comb. of 18 in 49. In actual fact, I had
to work with a sample as the entire population would have taken over 66
years to calculate, based on a two day test run, and I figured I didn't
have the time... It is likely that there was a 'better' set at the time.
Wouldn't it be lovely to have a Cray in the basement?
When calculating the winnings, remember that you are on top of a winning
pyramid because you have rolled out all the possible combinations. On one
draw, my numbers encompassed all 6 winning numbers plus the bonus.
Although, the 6 is a straight win and the 7th of no consequence to it, In
combination with 5 other tickets with 5 correct under them, the bonus
appreciably increased the winning pool.
Incidentally, that is the win pyramid. Given a winning combination of 6,
you have even more combinations of 5 under that, even more combinations
of 4 number winning tickets and still more 3 number winning tickets
because you have rolled all possible combinations.
I do play the lottery every once in a while, I like to dream as much as
the next guy even though a lottery ticket is about the closest thing to a
dead loss that you can get. If I was serious about winning and had the
money, I think I'd gamble on a neural net.
So are there any other crazies out there that burn up midnight oil on
such mad pursuits? What wild and crazy results issued from your fevered
imagination and specious incantations to your program?:-)
TTFN Jimi
------------------------------
End of FUZZY-MAIL Digest 157
****************************
FUZZY-MAIL Digest 158
Topics covered in this issue include:
1) Applying Advanced Technologies in Financial Services
by cecil@technet.sg (Cecil Lee)
2) Quantifying Tremor, fuzzy logic on 68HC11?
by rehrlich@ee.ryerson.ca (ROGER EHRLICH)
3) Open-ended sets in conclusion?
by palun@strix.udac.uu.se (Ulf Nordlund)
4) Re: backward chaining
by edd@csdc02.orl.mmc.com (Ed DeRouin)
5) Fuzzy Trafic
by wardda@tuns.ca
6) ISMVL'95: Multi-Valued Logic Symposium: Program/Registration
by "Jonathan W. Mills" <jwmills@cs.indiana.edu>
7) call for papers
by pha33@cc.keele.ac.uk (R. Rankin (Auto Id Grp))
8) CFP: Tools with AI '95, extended deadline
by bojan@cs.uh.edu ( Bojan Cukic )
9) help: how good a bird
by BARON RAWLINS <BMR116@psuvm.psu.edu>
10) Fuzzy sets and Concept Formation
by Roger Taylor <rt2p+@andrew.cmu.edu>
11) ANNOUNCE: FOOLpackage version 1.2 released
by Bernd.Landorff@arbi.informatik.uni-oldenburg.de (Bernd Landorff)
12) ASI-AA-95 (autonomous agents) -- final call
by dmeier@ifi.unizh.ch (Daniel Meier)
----------------------------------------------------------------------