Machine learning in trading: theory, models, practice and algo-trading - page 1271

New comment
 
Here the idea came to mind that the normal distribution can be expected only on completed processes (just a thought - didn't check), and the market is not, so all assumptions about its uneven distribution and non-stationarity are just guesses, which cannot be checked while there is a market.
 
Maxim Dmitrievsky:

For the market it is trained the same way, because the technology is the same. Forget about "influence", there is no such thing as explicit training in the way you imagine it. And there are sets of strategies, optimal strategies, etc. (in RL they are called policies).

If we had a bot creator, we could ask him a series of questions and be very surprised at the answers. All of these RLs are good at learning stationary environmental influences, but if you're playing against another player, that's where naked RL won't work. I haven't studied this area in depth, I may be mistaken. But there's a clear interaction with the changing environment, you can't see from the behavior of bots that there is an expectation of something, not at all, there is a control of probability of causing damage both to you and the opponent, and there are miscalculations simply and act towards a lower probability of a negative outcome, but it's not the probability given by policy at the beginning of the game - it's the effect on the changing environment.

 
I read you guys here. A couple of posts and realized that the theory goes on and on. But you will be very surprised when you start to practice. Building models and looking for new solutions is nothing compared to real trading, when you set everything up, set it up and wait ...... Then the first signal appears and the robot opens a trade. But the worst thing is when you analyze the result of this trade and its (robot's) decision. It will keep on making money or it's time to change it. In my experience I usually find that after optimization he is making one or three deals at a loss and then starts to rise. And at these moments, I also start to fuss and check. It will go up or not. It means that when you start real trading you face such problems that you didn't even think about when you were searching and optimizing. So the sooner you start to practice the faster you will face the problems of a practicing trader. IMHO of course
 
 
I watched the video and comments in Russian, and I was even surprised that I had a very correct idea of the behavior - there is an assessment of assets, there is a probability of outcome, there is a microcontrol, there are the initial strategies for development. There is a very great emphasis on the selection of units for fighting with the efficiency of micro-control and, apparently, the most effective from an economic point of view, and plus others are not needed, as the battles are won quickly. In general, clearly there are a number of components, and most likely not one neuron, and different models responsible for different components. Well, plus the developers themselves have stated that they use different models of basic strategies, which would be a kind of random.
 

Hi all. Since we have established that everyone here is quite intelligent and this is not sarcasm, but a statement of fact. In the field of MO comes people who are not lacking in intelligence, then I have a question for the community. Has anyone ever administered UBUNTU? I am talking about administration. The problem is this. During optimization this process is run as many times as there are cores in the system, paralleling calculations, etc.

Question: is it possible to force 2 or 4 cores to serve one running non-divisible process? Well, that's just me... maybe somebody knows about it...

 

Google works today

https://deepmind.com/blog/alphastar-mastering-real-time-strategy-game-starcraft-ii/

I got into a futile argument again yesterday

 
Maxim Dmitrievsky:

Google works today

https://deepmind.com/blog/alphastar-mastering-real-time-strategy-game-starcraft-ii/

Okay, I got into a futile argument again yesterday

Pay attention to the graph of expectation of the outcome of the battle, there is exactly the probability I was talking about - the situational probability that activates the feedback of neurons. You can clearly see how this probability changes as you scout and lose units on both sides - i.e. there is a constant recalculation of the balance of power and your expectations. This is clearly not a classic RL variant.


 
Aleksey Vyazmikin:

Pay attention to the graph of expectations of the outcome of the battle, there is just the probability I mentioned - the situational probability that activates the feedback of neurons. You can clearly see how this probability changes as you scout and lose units on both sides - i.e. there is a constant recalculation of the balance of power and your expectations. This is clearly not a classic RL variant.


So you're so clumsy that I thought that's what I was trying to tell you :D

Read about the Nash equilibrium in the link also, I described the algorithm 1 in 1 yesterday, without even reading

you wrote that no probabilities are estimated, but there is an Influence on the opponent :))) reread your posts

The game is a classic deep RL LOL, stop talking nonsense when you're not on the topic

 
Maxim Dmitrievsky:

So you're so clumsy that I thought that's what I was trying to explain to you :D

Read about the Nash equilibrium at the link also, I described the algorithm 1 in 1 yesterday, without even reading

you wrote that no probabilities are estimated, but there is an Influence on the opponent :))) reread your posts

The only difference is that you're not interested in the outcome.

This is a forum for trading, automated trading systems and testing trading strategies

Machine Learning in Trading: Theory and Practice (trading and not only)

Aleksey Vyazmikin, 2019.01.25 17:15

I see it slightly differently, in the toy conditionally there is a mathematical evaluation of each side, consisting of many factors - the number of bots and their potential, property, money, and the goal of the opponent to reduce this evaluation index so as to keep their evaluation index higher than the opponent, i.e. to spend less energy on the result. Thus you get a mutually influential system, where it is clear that by sacrificing a unit you will decrease the opponent's estimated asset value by more than the unit's estimated value, then this is the right decision, and if not, then it's wrong. And in trading we have no guarantees, only probability, but in a toy there are mathematical guarantees that can be calculated.

We can't influence the situation, but in the game we can, including creating advantageous situations ourselves.

Forum on trading, automated trading systems and trading strategies testing

Мachine learning in trading: Theory and practice (trading and not only)

Aleksey Vyazmikin, 2019.01.26 00:06

If we had a bot creator, we could ask him a number of questions and would be very surprised by the answers. All of these RLs are good at learning stationary environmental influences, but if you're playing against another player, then naked RL won't work here. I haven't studied this area in depth, I may be mistaken. But there is a clear interaction with the changing environment, from the behavior of bots do not see that there is an expectation of something, not at all, there is a control of the probability of damage to both you and the opponent, and there are miscalculations simply and act towards a lower probability of a negative outcome, but this is not the probability, set by the policy at the beginning of the game - it is the effect on the changing environment.


I do not know, maybe your brain is used to think with concepts from books, but it is easier for me to operate with less high matter, so I described in my own words.
1...126412651266126712681269127012711272127312741275127612771278...3699
New comment