Machine learning in trading: theory, models, practice and algo-trading - page 787
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
I just wanted to talk to someone, but no one but the wizard knows anything about it, so I'll talk to myself
:)))))))))))) It's mostly the kids around here, Maxim. And there's no need to quarrel with Koldun - you should read carefully what he writes. By the way, like my rampant opuses :)))))))
:)))))))))))) It's mostly the kids around here, Maxim. And you shouldn't quarrel with Koldun - you should read carefully what he writes. As well as my rampant opuses :)))))))
You're doing fine, Alexander, but I haven't understood half of it, so I can't talk about it yet :) And I have a poor understanding of distributions, how to compare and convert them, etc. I've just never done it
I will look into it later
RL is already a model with a teacher, but you don't teach it, and the outputs are selected randomly and then updated (this is called reinforcement learning)
what should i refute? overlearning can be both there and there, depending on the informativeness of the traits
In short, you seem to be intelligent people, but you write some nonsense, all the same so for 100 trips
I just wanted to discuss this method with someone, but other than the wizard no one knows anything about it
It's a re-learning process.
I understand you correctly that the problem of retraining is the same for all models and a separate, independent.
1. Do models with reinforcement solve the problem with the teacher himself, somehow automate the process?
2. Do reinforcement models not create additional problems in the area of overlearning? Well, at least from the fact that the teacher is matched to the plot as well?
I understand you correctly that the problem of retraining is the same for all models and separate, independent.
1. Do models with reinforcement solve the problem with the teacher himself, somehow automating the process?
2. Do reinforcement models not create additional problems in the area of overlearning? Well, at least from the fact that the teacher is being matched to the site as well?
I believe that independent
1. yes
2. can create miscommunication as the output labels will be constantly changing as they are taught and matched. But it is possible to analyze the dynamics of changing the labels and find the sets with the least amount of fitting
You're doing great, Alexander, but I don't understand half of it, so I can't talk about it yet :) and I don't understand much about distributions, how to compare them, convert them, etc. I've just never done it
I'll look into it later
80% correct inputs...
Well, you know, it depends.
I personally think that 100% is not enough. More precisely, such an estimate of the TS is not applicable at all in essence.
The TS should trade in the following way: any entry - buy/sell and always profit!
You're doing great, Alexander, but I don't understand half of it, so I can't talk about it yet :) and I don't understand much about distributions, how to compare them, convert them, etc. I've just never done it
I'll look into it later.
There's not much to read there. On the contrary, I'm reading this thread, waiting for a breakthrough in finding the neural grail. And I'm sure it's going to happen soon. Seriously. And I can even guess the name of the person who is about to literally blow us away with the results. But, I won't tell you. Let there be intrigue.
There's not much to read there. On the contrary, I am reading this thread, waiting for a breakthrough in finding the neural grail. And I'm sure it will happen soon. Seriously. And I can even guess the name of the person who is about to literally blow us away with the results. But, I won't tell you. Let there be intrigue.
As long as he's not too shy and doesn't leave the forum with all his bags stuffed :)
The main thing that would not silence then and did not run away with stuffed bags from the forum :)
That's for sure!
I believe that independent
1. yes
2. may create misunderstandings, because the output labels will constantly change during training and their fitting. But it is possible to analyze the dynamics of label changes and find the sets with the least fitting
the easiest way is early stopping, pre-learning. everything is the same as in normal teacher assisted learning
The important thing is not what will be in X bars, but what was in 10 bars, that is, if in 10 bars reached X points, then open.
Mihail Marchukajtes:
You're confused. I'm sorry. I sincerely ask you to adequately formulate your thoughts and arguments against, if any. You're absolutely right, we will look 10 bars back to get a prediction 10 bars later. This is how all TS-NSs are usually built.
We make the forecast first. We get its value and then we decide what actions to take at this value or that one....
Alexey is right. The forecast of the 5th bar may give +100 ppts, while the forecast of the 10th bar gives 0 ppts. As a result we miss a trade. Or the forecast on the 10th bar may show +100 pts. But on the 5th bar it may be -100 pts and you will catch a stop. It is necessary to forecast all of the bars from 1 to 10.
The disadvantage is increasing complexity of the network, because there are more exits and as a result the time of calculations is longer. Also it is necessary to increase number of neurons in hidden layers...
Read topkstarter's blog - there is also regression and a lot of smart ideas.