Machine learning in trading: theory, models, practice and algo-trading - page 346
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
BUT... the point of using this technology is not that it's cooler to do the job than humans, it's just convenience, speed, and high performance...
They're better at it, there's less recognition error.
highly specialized neural networks are better and much faster at recognizing images. Another thing is that if an object is outside the training sample and cannot be recognized correctly, a person connects his or her other brain resources, and these resources are simply not included in the NS
and they won't find anything directly "hidden", you just have to be conscious in your choice of predictors in my opinion
And what concerns LSTM, in my opinion, they must be good just for time series, because they are able not only to generalize signals for buying and selling, but also to adjust for the change of predictors' efficiency on different segments, depending on the market changes and the previous behavior of predictors, this will be the "memory" of NSThey do a cooler job, less recognition error
highly specialized neural networks recognize images better and much faster. Another thing is that if the object goes beyond the training sample and cannot be recognized correctly, a person connects his or her other brain resources, and these resources are simply not included in the NS
and they won't find anything directly "hidden", you just have to be conscious in your choice of predictors, in my opinion.
Ok... Well, have you ever tried to set a certain task which is obviously solvable but very difficult just to check what the network is really capable of doing experimentally...
For example a very noisy artificial sample with known deterministic constituent, some complex function, which if it was not known beforehand is impossible to calculate just by eye...
maybe there are some links to such studies....
One more thing: try a time lag network... it's the most suitable for predicting time series... I haven't tried it because I don't know much about it.By the way, what do you think neural networks are needed for in forecasting... I'll explain:
you need to understand the main thing - either neural networks are able to extract really hidden to the eye, complex relationships, or to automate some complex process, which cannot be easily formalized such as pattern recognition, which nevertheless can be easily handled by human brain/eyes, but there is a need to relieve a human from such routine tasks...
a simple example: face recognition at the airport. a person can do this task easily but it has become necessary to automate this process and this task cannot be solved by usual methods... some technologies like neural networks are connected... BUT... the point of using this technology is not that it is cooler than a person but just convenience, quickness and high performance...
what about this in trading: the task of pattern recognition using a neural network...why? can't the human brain solve this task? like face recognition...
out of 1000 people a person will recognize a familiar face in an instant...
do neural networks really have the potential to surpass biological neuronics...
NS has a potential that exceeds "human neuronics", at least in facial recognition and everything and anything). It's obvious, I think. Reduces human labor costs by tens or even hundreds of times.
Neurons, in general, are not for prediction. Prediction by neurons is more of a side effect of recognition (classification). Like: the sun is red at sunset (recognition) - tomorrow is a windy day (prediction).
A good trader usually trades with his hands better than any system, and the autosystem is not because it is better, but to avoid staring at the monitor and sitting in the toilet).
NS has a potential that exceeds "human neuronics", at least in facial recognition and everything and anything). It's obvious, isn't it? It reduces human labor costs by tens or even hundreds of times.
Neurons, in general, are not for prediction. Prediction by neurons is more of a side effect of recognition (classification). Like: the sun is red at sunset (recognition), tomorrow is a windy day (prediction).
A good trader usually trades with his hands better than any system, and autosystems do not do it better, but so as not to stare at the monitor and sit in the pants).
isn't this a contradiction....
Ok... well, have you tried to set a specific task which is known to be solvable but very complex just to check what the network is really capable of doing experimentally...
for example a very noisy artificial sample with a known deterministic component, some complex function, which if it was not known beforehand, it would be impossible to calculate it just by eye...
maybe there are some links to such studies....
I don't have much experience, right now I only have results of improving already working strategy with the help of the National System... that is, the National System, with the same predictors, has found better market entry signals than I tried to figure it out myself... this is a fact... I have no prejudices, I will look at the market later on.
I will read about time lag networks, thanks, I am not aware of it.
Isn't that a contradiction....
What do you see as a contradiction? I don't see it.
To the previous post. There are noise suppressors on neurons.
What do you see as a contradiction? I don't see it.
To the previous post. There are neuronics-based noise suppressors.
Well as in what...
NS has a potential exceeding the "human neuronics" at least in facial recognition and all that. But at the same time the trader manually, i.e. without AI, who has lower potential and who copes with all these tasks much worse why should trade better than any system ...
It's the purest contradiction.
and that Yuri, let's keep it on a first name basis, because it makes me laugh a little ... all these you pardon me...)
Well, as in what...
NS has the potential to exceed "human neuronics," at least in the same facial recognition and everything and everyone But at the same time manual trader, i.e. without AI, whose potential is lower and who copes with all these tasks much worse, should trade better than any system ...
it's the purest contradiction.
And this Yuri, let's speak on a first name basis, because it makes me a little nervous ... all these you pardon me...)
If you don't mind, I prefer to use the word "you".) Whatever you want, it doesn't bother me.
A sophisticated modern NS has about as much brains as a cockroach. A cockroach can successfully cope with quite a wide range of tasks. Due to the high speed of the NS, and the fact that the whole NS solves only one class of problems, it copes only with this class, and with nothing else.
Trader can see what NS has not been trained (or not learned)). Not only that, but unlike NS, he is able to cope with a wider range of tasks. That is, even theoretically, the trader has more possibilities, and he is able to trade better than NS, if it does not require superfast reaction.
By the way, NS are also mistaken. Probability of 90% is not a bad indicator for NS.
Such beautiful words about networks and philosophical reflections in general.
At the same time I do not see any evidence, not even a reflection on the topic of retraining nets
So all your networks are retrained or not?
PS.
That was the main question at the beginning of the thread.
There's RNN and RNN3 and its unpublished article on MQL5, and for nothing. RNN works better in optimizer than RNN3.
Because RNN3 has an error.
The data has been requested for 5 points, while the probability is calculated for 4 points and you don't start with the current bar a0, but with an offset for a period, i.e. from a1. Probability is calculated not for the current moment, but for a1 point - hence the results are bad for a0))
// Преобразование показаний технического осциллятора RSI в диапазон от 0 до 1 включительно
double a0 = rsi[0] / 100.0;
double a1 = rsi[p] / 100.0;
double a2 = rsi[2*p] / 100.0;
double a3 = rsi[3*p] / 100.0;
double a4 = rsi[4*p] / 100.0;
// Вычисление вероятности
double results=RNN(a1,a2,a3,a4);
i.e. we need =RNN(a0,a1,a2,a3);
Well, to count for 5 points - you need to increase the log matrix to 32 rules, i.e.25. And so on, if more points/inputs are needed. If there are 10 inputs, it means210 = 1024 input variables )))) I wonder if MT5 is able to optimize it?