Machine learning in trading: theory, models, practice and algo-trading - page 1680
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
What gifts, if you confuse the Witches with magicians)))
Classification - only with the teacher.
Clustering - with or without a teacher.
Any TS - predicts, the forecast horizon is equal
It's been written about a hundred times...
So sit with the cormorants)))
There is no need to quarrel, colleagues, over such a trifle as definitions. Simply speaking of theory, there are only two global directions.
The first tries to get a future result
The second is trying to get the current vectors to belong to one group or another.
Whereas in the first group the prediction is obvious and relates to a specific future, in the second group, the prediction may or may not be. If the target looks into the future, then the classification has a prognostic component, but it may not, if the target has no prognosis.
I'm just saying... to summarize, so to speak.
The spread is enough for everyone who earns on it, if it is not enough, then they involve "ingenuity" - requotes, large slippages - although the point is quite different - their earnings are guaranteed and risk-free, so to speak, pure brokerage
I'm not talking about it, it's not our lot or we have to drop everything and do it)))
If you believe the "most honest broker," then the spread is negative.) You can also think about rebates, bonuses, etc. This can be modeled by the game of minority from the game theory - the winner is always on the side of those traders who are in a minority. This game is a negative sum game for traders and on average they always lose.
Regarding the optimal TS that I mentioned, there is such a TS, banal - we open by the opening of the bar in turn by buy-sell-buy-sell, regardless of the outcome with a constant lot
with this TS is pretty quickly found the optimal StopLoss and TakeProfit, but this TS works more confidently only with increasing TF - I checked a long time ago, it seems on the daily bars is more or less stable
BUT... and there will be problems - crises and changes in rates of the Federal Reserve System cause a series of defeats/victories in such TS
i.e., i think that even to find the TS "constantly dangling" near zero, most likely, is not real to find - tomorrow i will check GA tester
I do not know a TS that can work without modification (or optimization) for different instruments, ie the problem is the same - to find out that the current instrument has stopped TS working, but not to drain the deposit, and what to do in the future - to seek another TS or another instrument - is an afterthought
I was talking about the equilibrium (in the sense of Nash) solution to the game. It's quite strange from the real point of view (but quite simple from the abstract mathematical one), when all traders make a decision about the direction of entry, simply by flipping a coin. The main question here is how the price behaves - two extreme possibilities are possible:
1) the market also flips a coin without peeking at traders' decisions and then the game will be zero-sum
2) the market looks at the traders' decisions and simply moves the price in the direction that is profitable to their minority and leaves for itself the difference between their winnings and the losses of the majority.
For all its conventionality, the second variant seems to me more truthful. Of course, there is no practical use in this. Except that you can conclude that the grail is impossible)
The difference is that in the first case we try to assign the input vector to this or that class taken from the target class. In the second case we try to divide vectors into groups depending on distances of these vectors (conditionally). Just methods are different but the essence is the same. That is why clustering is a subspecies of classification. Questions???
Wrong at all. You are mistaken.
I will give you examples when the time comes.
So Max, the point of reference is that there is Classification and Forecasting and all methods can be referred to one or another group fundamentally and there is no third one at the moment. As soon as I come up with one, I'll be sure to let you know, but for now sleep well. We're all right, just each in his own way...
Well, to be precise, there is Classification and Regression. The result of the first one is a prediction and the second one is a forecast withconfidence interval.
Well, to be precise, there is Classification and Regression. The result of the first one is a prediction, while the second one is a forecast with a confidence interval.
Prediction and prediction are synonyms, so...
Misha! Read a book, don't embarrass yourself.
Misha! Read a book, don't embarrass yourself.
Cowards invented books :-) Just kidding. One way or another, the whole conflict is precisely in the concept of definitions.
All people want happiness, but few can agree with each other just because of different concepts and different depth of understanding.
They tell you that classification is learning without a teacher, but you say no! Classification is when you have a tummy ache ... so you have a conflict, because you do not understand the subject or the concepts, so I say, read a book!
Or I say that prices need to be preprocessed, and the DSP does it, and they say to me it's for radio amateurs) Well, what's there to talk about if they're wooden at least up to their waists ... Conflict))