Discussion of article "Metamodels in machine learning and trading: Original timing of trading orders" - page 5
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 can still give you advice, if you want it for money, I can give it to you for free.
For this topic, there can be advice on the part of feature selection (for example, on information criterion). And show that these features are better than increments. Psychology and Higgs boson as signs do not work.
For this topic, there may be advice on how to select features (e.g., information criterion). And to show that these signs are better than increments. Psychology and Higgs boson as attributes do not work.
I did not write anything about the boson. Valeriy Yastremskiy thinks that the metamodel should be more complex. I completely agree with him.
The task of checking indicator readings on a wider range and comparing them with readings on a narrow range is interesting, but what do the indicators themselves show?
Indicators should provide information about the market state. In my understanding, this state should be described by the probability of upward or downward movement, which is calculated as the Hamiltonian of the system. The kinetic energy of the market is quite simple to calculate - the speed of price movement multiplied by the momentum. Potential energy is more complicated, but it is also solvable. The problem is a very large amount of work that must be done to bring all the formulas and figures into one working programme. And the work is qualified, implying fluent knowledge of maths, physics and programming.
Some idea about visualisation of fields is given by the attached picture.
I did not write anything about the boson. Valeriy Yastremskiy thinks that the metamodel should be more complicated. I completely agree with him.
The task of checking indicator readings on a wider range and comparing them with readings on a narrow range is interesting, but what do the indicators themselves show?
Indicators should provide information about the market state. In my understanding, this state should be described by the probability of upward or downward movement, which is calculated as the Hamiltonian of the system. The kinetic energy of the market is quite simple to calculate - the speed of price movement multiplied by the momentum. Potential energy is more complicated, but it is also solvable. The problem is a very large amount of work that must be done to bring all the formulas and figures into one working programme. And the work is qualified, implying fluent knowledge of maths, physics and programming.
Some idea about visualisation of fields is given by the attached picture.
The increments give on average 0.01 (out of max 1) according to the criterion of mutual information with labels. That is, there is almost no correlation, random.
There are augmented increments of the type
which give more. The maximum information is carried by the bare graph, but it is not digested by neural networks and boustings. The task is to bring the series to stationary with minimal loss of information.
There is nothing else to catch there, from the word "absolutely" by any Hamiltonians and so on.
This is a question that must be solved before you want to complicate the metamodel, because no model can be trained on randomness.
The increments give an average of 0.01 (out of max 1) on the criterion of mutual information with labels. There are augmented increments of the type
which give more. The maximum information is carried by the bare graph, but it is not digested by neural networks and boustings. The task is to bring the series to stationary with minimal loss of information.
And for dummies - labels of what? increments of what? What initial data are processed?
For dummies - labels of what? increments of what? What raw data are processed?
Labels are trades, for buying and selling. We need an information link between attributes (price increments, for example) and the predicted direction of deals. The task of the proposed algorithm is to search for this relationship.
This is the basics of econometrics.marks are deals, for buying and selling. We need an information link between the signs (price increments, for example) and the predicted direction of deals. The task of the proposed algorithm is to search for this relationship.
This is the basics of econometrics.And what principle is used to open deals? At random, or by some indicator, or by clock, or another algorithm?
What principle is used to open trades? At random, or by some indicator, or by clock, or another algorithm?
it doesn't matter, you can search for dependence through oversampling. In the previous articles, random sampling of trades with correction of unprofitable ones was suggested.
There is a set of deals and a set of signs, we need to find signs predicting the direction of these deals
The algorithm in this article throws out bad pairs of trait-trade pairs, leaving the most predictable ones (by information criterion).
it does not matter, one can search for dependence through overshooting. In the previous articles random sampling of trades with correction of unprofitable ones was proposed.
There is a set of trades and a set of signs, we need to find the signs predicting the direction of these trades
The algorithm in this article throws out bad pairs of trait-trade pairs, leaving the most predictable ones (by information criterion).
And how is this method better than astrology or fortune-telling on a ram's liver?
And how is this method better than astrology or divination on a ram's liver?
it is difficult to communicate with nubbies, the article is for those more or less trained in MO.
it is difficult to communicate with nerds, the article is for the more or less trained in MO.
And I naively thought that neural networks should help the thinking process when making decisions, not dumb it down.