Machine learning in trading: theory, models, practice and algo-trading - page 2726
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 did about all this, I guess the error is in another place, I'm debugging further...
thanks for the advice
However, it is better to refer to the sections Market, Signals and Freelancing. There are important theoretical issues that need to be discussed constantly, and it is impossible to do it anywhere but the forum.
Personally, I am still very much interested in the question of constructing an algorithm for determining the length of a historical sample for training. The "let's take N years (months, days)" method doesn't seem quite appropriate. Why not N+1, for example?
However, it is better to refer to the sections Market, Signals and Freelancing. There are important theoretical issues that need to be discussed constantly, and it is impossible to do it anywhere but the forum.
Personally, I am still very much interested in the question of constructing an algorithm for determining the length of a historical sample for training. The "let's take N years (months, days)" method doesn't seem quite appropriate. Why not N+1, for example?
The answer is in the theory you follow:
1. The market is changing
2. The market is not changing
Variability refers to identified probabilistic patterns.
In the first case, such a segment should be searched for, and in the second case, one should take everything that is available.
I adhere to the second approach, that's why the change of MOEX trading session just kills me, and I lose the desire to engage in MOEX.
But again, I take quotes from 2014, as the market changed significantly in its volume and movement then.However, it is better to refer to the sections Market, Signals and Freelancing. There are important theoretical issues that need to be discussed constantly, and it is impossible to do it anywhere but the forum.
Personally, I am still very much interested in the question of constructing an algorithm for determining the length of a historical sample for training. The "let's take N years (months, days)" method doesn't seem quite appropriate. Why not N+1, for example?
The easiest way to understand it is to superimpose the charts of the instrument and the balance of the TS. Then you can see when it breaks and why, usually the breakdowns of the TS come at the change of the trend or when prices go beyond the ranges on which it was trained.
The answer is in the theory you hold:
1. The market is changing
2. The market is not changing
Perhaps my theory is closer to the first point - the market can change sometimes, although it doesn't do so in any regular way. Otherwise I wouldn't be asking this question.
The easiest way to understand it is to superimpose the charts of the instrument and the balance of the TS. Then you can see when it breaks and why
We get a posteriori analysis of an already trained model. I would like to supplement it with a priori analysis for the stage of selecting a training sample.
Usually TC breakdowns occur at trend changes or when prices go beyond the ranges on which the model was trained.
I think so too. I stopped using the last formed top of the zigzag for simplicity, but I would like something more elaborate.
Perhaps my theory is closer to the first point - the market can change sometimes, although it doesn't do so in any regular way. Otherwise I wouldn't be asking this question.
Then you have to learn to predict similar market phases, though no, you have to learn to predict how the market is likely to change.
If every trend is not similar to a new trend, then this is the only way.
I rather think that there are several different forms of trends in trend and flat, and they do not change so much.
Probably it can be checked in some way, if you make an adequate markup by cutting the chart into trends.
Then you have to learn to predict similar market phases, though no, you have to learn to predict how the market is likely to change.
If every trend is not similar to a new trend, this is the only way.
I rather think that there are several different forms of trends in trend and flat, and they do not change that much.
Probably it can be checked somehow, if you make an adequate markup by cutting the chart into trends.
Your assumptions seem too strong. In the sense that if it were possible to realise them, it would be almost a grail. I would like to solve a more modest and specific problem - to find some general way to find a compromise between sufficient length of a trayne and absence of obsolete examples in it.
In my opinion, this issue is fundamental for applications of MO and matstat in our field.
to find some general way to find a compromise between sufficient length of the trayne and absence of obsolete examples in it.
We can also look from the often expressed point of view "we should not try to predict the market in the future, but to determine its state in the present". We need a meaningful way to identify this very "present". Moreover, there can be several such "present" (different scales of the "present") - the main thing is that there should not be too many of them and that the selection of each one should be meaningful.
Your assumptions seem too strong. In the sense that if it were possible to implement them, it would be practically a grail. I would like to solve a more modest and specific problem - to find some general way to find a compromise between sufficient length of the trayne and absence of obsolete examples in it.
In my opinion, this issue is fundamental for applications of MO and matstat in our field.
Have you not tried experimentally? After all, according to your theoretical approach to this question, after a critical increase in the sample size, the patterns in the sample will be old, no longer working, which means that learning should deteriorate in a qualitative sense and the results on new data will be worse when the sample is increased.