Just a little intro about myself. I'm working on commercial spam filtering software which utilizes neural networks along with statistical methods. The neural...
2146
Mattias Fagerlund
mattias_fage...
Aug 3, 2005 8:01 am
Hi! That's an interesting read, but their ratio of false positives seems extremely high. But if it's used as one indicator among many, I guess that's...
2147
logical_creations
logical_crea...
Aug 3, 2005 8:44 am
Hi to the board. My name's Chris Turner. I'm a software developer in England and I'm interested in experimenting with Neural Networks and how they can be used...
2148
Chris Turner
logical_crea...
Aug 3, 2005 11:11 am
... I managed to solve the problem. I made the assumption that as the output signal is between 0 and 1, that the input signals should also be beween 0 and 1....
2149
Matt Simmerson
easablade
Aug 3, 2005 12:08 pm
Hi Just a quick intro to myself. I have been 'playing' around with NN's and GA for a few years now and came accross NEAT through Mat Buckland's excellent...
2150
Chris Turner
logical_crea...
Aug 3, 2005 4:20 pm
Hi, I have a couple of questions that I'd like to put to the group. I suppose these questions may have been asked before but I've spend sometime searching...
2151
Alexander N. Telegin
since7bb
Aug 3, 2005 4:26 pm
Yes, that's why I'm going to try NN along with statistical filter. The result will be summarized according to weights for each method. Regards, Alex...
2152
Philip Tucker
tucker0171
Aug 3, 2005 4:31 pm
... Is #11 the bias input?...
2153
Kenneth Stanley
kenstanley01
Aug 3, 2005 7:14 pm
Maybe Colin (the author of SharpNEAT) can explain this...I'm not sure about the inner workings of SharpNEAT. ken ... resets ... 89 ... is ... wrong ... also...
2154
Kenneth Stanley
kenstanley01
Aug 3, 2005 7:21 pm
Hi Matt, fitness sharing is meant to penalize a species the larger it gets. Of course, species only tend to get very large if they have high fitness, which is...
2155
Kenneth Stanley
kenstanley01
Aug 3, 2005 7:25 pm
Chris, How many generations have you run your experiment? Sometimes it can take a while to build up the necessary structure to jump from "almost perfect" to...
2156
Colin Green
alienseedpod
Aug 3, 2005 7:54 pm
... Hi Chris, As you probably noticed most of the SharpNEAT experiments scale inputs to the 0..1 range as you did, although I generally consider this to be not...
2157
Colin Green
alienseedpod
Aug 3, 2005 8:27 pm
... Philip got it. There is always one bias input node + however many you define for your experiment, and it is always the first node. ... Can you explain the...
2158
Colin Green
alienseedpod
Aug 3, 2005 8:40 pm
... If each technique reqults in some sort of confidence value then it might beneficial to evolve a network that accpets these confidence values as inputs...
2159
Alexander N. Telegin
since7bb
Aug 3, 2005 9:12 pm
... might ... values as ... may be ... different ... then 2 ... may be ... relationships ... Yes, interesting idea. In my opinion, such summation may have ...
2160
Colin Green
alienseedpod
Aug 3, 2005 9:33 pm
I got this working over the weekend, results so far are not brilliant, but interesting. Keep in mind that I'm still evaluating against a dominance chain which ...
2161
Colin Green
alienseedpod
Aug 3, 2005 9:38 pm
Colin Green wrote: Correction, the overall lose rate is 8% not 16%......
2162
Chris Turner
logical_crea...
Aug 4, 2005 7:29 am
Thanks to everyone for the replies. I've only been playing around with NEAT for a couple of days and I've staill got a lot to learn. I've picked up a few idea...
2163
Matt Simmerson
easablade
Aug 4, 2005 9:40 am
Hi Ken Thanks for that. The fog is now beginning to lift. Yes I do intend on making my version of NEAT available. It is part of a much larger AI library so I...
2164
Ian Badcoe
ian_badcoe
Aug 4, 2005 11:13 am
At 22:32 03/08/2005, you wrote:
>I got this working over the weekend, results so far are not brilliant,
>but interesting.
>
>Keep in mind that I'm still...
2165
neat@yahoogroups.com
Aug 4, 2005 11:28 am
Hello, This email message is a notification to let you know that a file has been uploaded to the Files area of the neat group. File : /TTT Whole...
2166
Ian Badcoe
ian_badcoe
Aug 4, 2005 11:32 am
Hi,
In case it is useful to anybody, I uploaded a text file with a complete analysis of TTT using normalisation to reduce the number of boards to a...
2167
Ian Badcoe
ian_badcoe
Aug 4, 2005 12:45 pm
Oh, and importantly for anybody working on this, I can give you the code that produced it. It's C++ but shouldn't be too hard to convert to whatever. And...
2168
Philip Tucker
tucker0171
Aug 4, 2005 1:31 pm
... We found that the traditional "perfect player" basically just plays defensively and makes 3-in-a-row when he stumbles across it. But, he doesn't fully...
2169
Ian Badcoe
ian_badcoe
Aug 4, 2005 2:34 pm
At 14:31 04/08/2005, you wrote:
> > _IS_ it ever correct to count wins in TTT? e.g. because you cannot win
> > against a perfect player, the best you can do...
2170
Ian Badcoe
ian_badcoe
Aug 4, 2005 2:58 pm
2171
Kenneth Stanley
kenstanley01
Aug 4, 2005 5:10 pm
Hi Matt, sure if you get together a complete version with documentation of some sort (doesn't have to be fancy) I will put it up on my site. Of course, there...
2172
Derek James
blue5432
Aug 4, 2005 8:58 pm
... Well no, that's not true, as you seemed to point out in the previous part of this post. If you're goal is to win, whether you're playing a single game or...
2173
Colin Green
alienseedpod
Aug 4, 2005 9:12 pm
... No, it is possible because the scores are weighted. A bit more info... To obtain a fitness score a new player is tested against each player in the...
2174
Colin Green
alienseedpod
Aug 4, 2005 9:48 pm
... This is an important aspect to consider, especially going forward looking at other more complex games. In fact it's absolutely crucial for some (most?)...
