Drake7707

Like building a large hadron collider with bombs & ships ? All aboard the particle stream !

Ah I don't mean a group, i mean like an official fan page (the "become a fan" feature".) Which acts more like a person, and can have status updates etc, a group can't do anything of this You can sync twitter updates to facebook and myspace too, which would mean multiple updates to these sites could be made in one place. http://www.facebook.com/pages/ContinuumSubspace/20806916745?sid=3fd341085cb5abe3aa6eb9d00fad6dcc&ref=search ?

Well it seems that 20x20 images with Tahoma 11pt ('A', 'B', 'C' or 'D'), and random noise pixels stabilises the network with 400 input neurons (obviously), 200 hidden neurons and 4 output neurons. If i increase the available characters to the entire alphabet the accuracy of 1/3 of the characters doesn't seem to be improving after a few training sessions. Apparently 200 hidden neurons can't contain enough information about all 26 characters. I wonder how the amount input & output neurons should influence the amount of hidden neurons :/

But the benefits are that half of his programming genes have been ported ^^

If I may believe my slides that explained the topic somewhat, the backpropagation algorithm also works with the treshold function, but instead of using the derative of the scaling function, 1 is used (like the term for the derative is ommitted). If i could find a good sample implementation that's well documented, I would at least be certain that I'm implementing it right and have a bug somewhere, but right now i'm not even sure that i'm using the correct formula's :/ Edit: Well i'm finally getting some promising results , i made a quick character recognition demo, reverted back to the sigmoid function, fixed the formula (* instead of + >.>), made sure that the weights are random between [-1,1] rather than [0,1], and some optimizations here and there, and it seems to be working

wouldn't Nigeria fit better?

Milo with Project Natal looks very good too.

marker with open tip on a desk

Tales of Monkey Island!

How did you solve the problem with the fixed child menu ? Did you put the style to none and made a custom control box for maximizing/restoring/minimizing/closing ?

There are some facebook groups for subspace/continuum. Everyone seems to be on facebook nowadays and each one has probably old friends that are/were in their msn friend list from continuum that were added on facebook using the friend finder, so if one group actually stands out and friends invite their friends to join the group you can create a cascading effect to reach much more people that used to play the game. (I should probably posted this in the ideas for advertisment or whatever that thread was called thread)

How can you use a monochrome bitmap for a selection mask ? You have to know the tile values that are selected without looking at the source (otherwise selection isn't a different layer anymore, like in the old days of DCME, where I used rectangle with reference to the source tiles)

It wouldn't be that hard to 'pin' the map and region layer together when you move stuff. I'm sure samapico will fix that up in less than a minute. I'll even make an icon for the toolbars ^^

doubt that anyone will care but i think i traced the error back to the backpropagation algorithm. After trying to use XOR instead of an AND gate (which is less head-on correct with random weights), i switched the error calculation to error = desired - actual and I also used the treshold function ( if >= 0.5f then 1 else 0) instead of the sigmoid function. I also experimented with leaving out the bias that is included in each neuron, but both scenario's still go haywire. One of the problems I encountered was that the weights went up to a gazillion, so i made sure to clip them in the interval [0,1]. If anyone has or seen a decent backpropagation tutorial, do tell, because i'm stuck again

but but but ... my precious icon ? No seriously, mine sucks

Nice, but you have to watch out for rounding errors here, if you add up a lot of small delta's the sum of the rounding errors could get pretty nasty which lead to the inaccuracies you speak of

dammmnnuuu I've seen it in my course Knowledgebase systems, but only theory, not an implementation. I've made an assisting document with documentation about the structure, it'll be useful to get a better understanding of the code, I attached it with newer code NeuralNetwork.zip

Congrats ^^, I'll be getting my diploma next month, after my presentation of my internship

Is anyone familiar with writing a neural network ? I was trying to develop one in C# but it appears that it's not working well (i think). I was trying a simple example of letting it behave like an AND gate (if i inputted 1 & 1 it would give a result close to 1, otherwise below 0.5), and I used back propagation to let it change the weights between the neurons to let it learn. Butttt apparently it doesn't converge to the desired result. Any suggestions if you know something about it NeuralNetwork.rar

Weird :/, normally a global error handling is in place that at least gives more information about the exception

Warning! Beware: Unstable release. May or may not cause nuclear fallout on crash

Updates wooooooooooooooooooo

I think that religion in its basic concept is nothing more than beliefs for people, to get them through difficult times, to share opinions with others that agree, to explain things that are hard to understand or to find solutions for it, etc. The reason why people think religion is bad (and probably why this topic exists) is because a minority of people force beliefs onto others, saying that they are right and all others are wrong and forcing people to do something leaves a bad and hostile impression. There are people like that everywhere, in every religion or culture, you even encounter them as a collegue or boss, an annoying friend you shouldn't really call 'friend' and so on. There have been persons or groups like that in the past (crusades, nazi's, ...), there are people like that now, and there will probably be in he future.

i got it too, a while ago. It's not really that hard to make a bot for that, I could whip one up in an hour or so (unless it asks for captcha's , then i'd need to do some neural network training first) Not that i would make one, it's quite pointless ^^

'Inkheart', 'Tale of a desperaux' and 'Bedtime stories' were pretty good ^^