Machine learning in trading: theory, models, practice and algo-trading - page 1787
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 don't use naked increments - just relative normalized measures essentially.
It makes no sense to mix predictors for model performance and predictors for determining the favorability of applying a particular model. I think one model should determine the favorability and the other the TC itself. Then there remains the question of markup for the training of favorable conditions, and for this it is necessary to define the threshold when the TS works effectively. This may be some indicators, for example the balance of errors and profit growth, or maybe some other metric indicators. And, consequently, the classification should probably be for a week or at least for a day.
Of course, the relative indicators should be. In the bare scale will have to be taken into account))) Relativity is more correct than difference.)))
The objectives of the significant characteristics of the series and the TC settings are of course different and you can not mix them. But they are interrelated. Bad or good TC settings for a given series. The cycle and randomness. We find the necessary condition and the optimum TS. But it does not mean that this is the best combination. And the further search for the optimal series may not coincide with the previous one))))
It is better to do the question of indicators from the opposite direction. From the maximum average and drain) In relative units. We are not looking for the threshold of efficiency, but the areas of work and plum.
Classification should be on all data. The notion of working on minutes or on the hour is wrong in essence. Work in the moment. This analysis is for the 1-minute or hourly timeframe, which to me is a rough assumption for making a decision and a source of errors.
There is no riddle here.
1) It is necessary to determine favorable periods for the TS and unfavorable periods, i.e. to create the target "Y" in the same usual binary form Y = 0000111100000
2) Create variables, which will reflect "market characteristics" 2. Create variables that will reflect the "market characteristic", honest and not displaced. The DSP, in particular spectral analysis, will help here.
From DSP we know that a signal of any complexity can be described by the sum of sine waves, a sine wave has only three parameters - amplitude, frequency and phase; this sum of sine waves or rather their parameters can be taken as a market characteristic and it will be objective.
If it is difficult for you, you can prepare the data for me, the price and "Y" for the classification, and I will make up a code and check if it is possible to detect favorable conditions for trading, because this topic is interesting for me, too.
The periods themselves do not give us anything. It is necessary to define the meaningful characteristics of these periods by which they can be classified.
There must be some logic in the variables. If you don't have it, you can't detect error. Without logic, you can only do it empirically and there is certainly a probability, but it is small. Sinusoids are useful when you know what they mean.
On the issue of learning from current data, it should be by criterion, the data can be classified by history or not. You need to compare the results of learning the significant characteristics of the series on history and current, if the combination is new, then the risk of errors increases.
The periods themselves provide nothing. We need to identify meaningful characteristics of these periods that can be used to classify them.
There must be logic in the variables. If there is no logic, the error cannot be detected. Without logic, you can only do it empirically, and the probability is certainly there, but small. Sinusoids are useful when you know what they mean.
On the issue of learning from current data, it should be by criterion, the data can be classified by history or not. It is necessary to compare the results of learning the significant characteristics of the series on the history and current data, if the combination is new, then the error risks increase.
What do periods have to do with it, did I say that? You don't understand anything, the AFR (amplitude-frequency response) is an objective characteristic of the function, in this case the market.
There is no riddle here.
1) It is necessary to determine favorable periods for the TS and unfavorable periods, i.e. to create the target "Y" in the same usual binary form Y = 0000111100000
2) Create variables, which will reflect "market characteristics" 2. Create variables that will reflect the "market characteristic", honest and not displaced. The DSP, in particular spectral analysis, will help here.
From DSP we know that a signal of any complexity can be described by the sum of sine waves, a sine wave has only three parameters - amplitude, frequency and phase; this sum of sine waves or rather their parameters can be taken as a market characteristic and it will be objective.
If it is difficult for you, you can prepare the data for me, the price and "Y" for the classification, and I will make up a code and check if I can distinguish favorable conditions for trading, because this topic is interesting to me, too
Thanks for the willingness to take part in solving the puzzle!
What about the indicator, about which you asked me earlier - do you have a specific TOR? Calculations all appear to be possible to do - gave links to the library.
If I look at the implementation of the idea of classifying sections, I think there will be some fixed timeframes within which the forecast will be given, it is necessary so that there would be no retraining and adjustment, otherwise if I give a forecast at every minute bar, there will be a lot of unnecessary noise.
As for the marking, it is not clear yet - we need to understand the criterion, or rather to experiment with different criteria for assessing the quality.
I marked myself the implementation of this idea in the plan of the ATC development - goes to number 17 :) Therefore, I am not sure that the problem can be solved quickly - we must decide how to solve it and how to check the result.
Maybe to make a tool on the basis of DSP, that you plan to apply, on MT5 and already here to see what comes out?
What do periods have to do with it, did I say that? You do not understand anything, the AFC (amplitude-frequency response) is an objective characteristic of the function, in this case the market
1) It is necessary to determine the favorable periods for the TS and unfavorable periods, i.e. to create the target "Y" in the usual binary form Y = 0000111100000
2) Create variables, which will reflect "market characteristics" DSP, in particular spectral analysis, will help here.
In general, I am not against the AFC series. If it will be possible to relate it to good and not good periods. Just the AFR is not always an objective characteristic of the series in relation to other characteristics we need. Decomposition is not a problem, the problem is to find the connection.
Of course the relative indicators must be. The bare scale would have to be taken into account.)) Relativity is more correct than the difference.)))
The objectives of the significant characteristics of the series and the TC setting are of course different and you can not interfere with them. But they are interrelated. Bad or good TC settings for a given series. The cycle and randomness. We find the necessary condition and the optimum TS. But it does not mean that this is the best combination. And the further search for the optimal series may not coincide with the previous one))))
It is better to do the question of indicators from the opposite direction. From the maximum average and drain) In relative units. We are not looking for the threshold of efficiency, but the areas of work and plum.
Classification should be on all data. The notion of working on minutes or on the hour is wrong in essence. Work in the moment. It is the analysis for the minute or hourly timeframe that, to my mind, is a rough assumption for decision making and a source of errors.
You can analyze a minute as well - I just want the prediction to be for a long period, or before the event/situation happens/changes. I don't think it makes sense to catch one micro-trend within it, and to accumulate stops on reversals.
Only how to count Y? just on the profit is probably not the best option, the entry point is important ... After all, the profit was obtained from a good entry point, not the range between entry and exit.
So it turns out we need only the entry point of the system and the market parameters at this time ...
It turns out that AMO will receive a signal from TS to enter and decide whether to open a position or not
It's scary to think but that's what our Micha constantly trended))
This is the question - whether we need entry points or the range... This is the question, whether we need entry points or a range. I would not want to be tied to entry points, because not every TS is easy to determine the entry points, and there may be lots of exit points - we need to determine the TS for the participants, assuming that it is effective over the whole favorable area.
It is possible to analyze minutes - I just want the prediction to be for a long period, or until the event/situation changes. For example, we predicted that most likely today will be a flat day, there is no point in catching one micro trend within it and collecting stops on reversals.
The analysis should be on every tick. It is impossible to do without it. Gaps and other nonsense. The forecast can only be on historical data, so it is important when the system does not recognize the current situation (comparing historical data and received data) and it is not a trained, not memorized by the system situation.
This is the question of whether you need entry points or still a range... I would not want to be tied to entry points, because not every TS is easy to determine the entry points, and exit points can be set - we should determine the TS on the participants, assuming that it is effective in all favorable area.
And range and entry points. Expectation decreases separately. ))))
1) It is necessary to define favorable periods for TC and unfavorable periods, i.e. create target "Y" in the same usual binary form Y = 0000111100000
2) Create variables, which will reflect "market characteristics" 2. Create variables that will reflect the "market characteristic", honest and not displaced. DSP, in particular spectral analysis, will help here.
Ahh yes, sorry my bad idea, I didn't mean the periods of oscillation or something like that, but just portions,favorable areas were supposed to be
Just frequency response is not always an objective characteristic.
Actually it always is.)
Decomposition is not the problem, the problem is to find the connection.
without experiment it's lyric.
That's the question - whether you need the entry points or still the range... I would not want to be tied to entry points, because not every TS is easy to determine the entry points, and exit points may be set - we need to determine the TS on the participants, assuming that it is effective in all favorable area.
I don't know, think, try, for me the entry point is easier and more objective, but it's imho
Ahh yes sorry my mistake, badly expressed, by periods I did not mean the fluctuations or something like that, but just areas, favorable areas were supposed to be
Actually always)
without the experiment is lyric.
To look at it is easy, select the sections and see how these sections differ in all the time points and you need to catch the differences.