Artificial neural networks. - page 11

 
A100:
The result is better than any criticism. The wrong methods were chosen to solve the problem in the first place
Can you enlighten us more, or are we going to play the stern guru?
 
A100:
Does your neural network want to break me for a few quid? There's a tradition here of betting on the results of championships.
 

I deleted all my posts just in case, the moderator thinks it's advertising

 
A100:

I deleted all my posts just in case, the moderator thinks it's advertising

can you delete your account too? :)
 
sergeev:
can you do your account too? :)
Whatever you say, dear moderator.
 
A100:
As you say, dear moderator.

- Scientists recently crossed a fly with a fly swatter!!!

- So?

- A very suicidal specimen.

 
Urain:

- Scientists recently crossed a fly with a fly swatter!!!

- So?

- A very suicidal specimen.

The breeders bred, and the environmental scientists didn't have time to put the samoyed animal in the Red Book...

;)

 
MetaDriver:

Right. And chess programmes will never learn how to play above second rate.

I've heard that before.

--

Vladimir, I hope your insanity is temporary, and I wouldn't want it to last for thirty years (like Marvin Minsky's).

But it's fun, yeah.

;)

Great idea, by the way!

That's when neural networks will be able to learn how to play (just the rules) chess, looking at the games they've already lost before, and I'll believe that with such a primitive tool as neural networks one can do something decent.

Neural networks are primarily designed for pattern recognition (automatic search for previously seen situations) but not for detection of any patterns.

 
papaklass:

So maybe traders are not teaching the network properly? How is it going now (in my layman's opinion):

1. the trader chooses the time frame in which the training will take place.

2. Selects the input signals (indicators, bar prices, etc.).



I will use your post to lean into the answer and join one of the previous authors - The best neural network is a Fourier series decomposition! Take the whole story, take a short waving scale of, say, 10 minutes, decompose it into a series, get a "time machine" hidden behind the coefficients, use it to "predict" the future in a tester, and you're done. You've got a simply genius result on the whole story. But at least here you can see this time machine, but in neural networks you can hardly see it. Neural networks can recognize images; images can be anything; there are more efficient methods of recognizing images, but what makes neural networks good is that they can retrain according to the fact of what is going on now. That's their thing. But that's also the ONLY thing about them. So to mentally analyse the applicability of neural networks, imagine that it is simply a pattern recognition system.

 
SProgrammer:

Great idea by the way!

When neural networks will be able to learn how to play (just the rules) a game of chess, looking at the games already lost before, I will believe that with such a primitive tool as neural networks one can do something decent.

Neural networks are primarily designed for pattern recognition (automatic search for previously seen situations) but not for detection of any patterns.

I've read that there are NSs successfully playing checkers at a high level. The case for chess won't be up in the air for long. I think it's more than possible. All these games refer to games with complete information, which means that uncertainty is only present in the opponent's progress. A probabilistic approach will allow you to look for better moves. I think probabilistic NS would be suitable for tasks like chess.