Machine learning in trading: theory, models, practice and algo-trading - page 856
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
For me? I've already solved the problem. Now I'm thinking about what else to do. Python or R. I have no new ideas yet.
You have to learn both, but remember, that only R has a reliable and reliable gateway to MQL.
Go to the next level - keras/tensorflow/. There are so many ideas there, I wish I had enough knowledge and time to master them.
The only thing you need is to have enough knowledge and time to master it. Good luck,
Check it out, I got it out of contact. Very useful information as part of understanding the market!!!
Bifurcation point.
There is a special concept in thermodynamics that can be adapted to almost any complex dynamic system. From time to time any such system, be it a state, economy or human psyche, enters a critical state of uncertainty.
At this point the orderliness of the system is threatened and its further development can follow two possible scenarios: either disintegration to a chaotic state or reaching a qualitatively new level of orderliness. For example, the bifurcation point for the state can be called a period of political instability, for the economy - an economic crisis, and for a person - a traumatic event.
In management theory three types of systems are considered:
Non-deterministic systems are systems which at some points in time can behave as deterministic (people march to the march "Farewell to Slavyanka") or random, for example the flow of people in the subway: everything is random, but well described by the theory of mass service. But if some perturbation is introduced into this crowd (Bomb!), then further behavior of all these people has nothing to do with the previous one.
One of the signs of indeterminate systems is the participation of humans in them.
In the late 1960s my university, in the department of Automatics and Telemechanics, had two considerably different specialties: automated systems (8 groups) and automated systems (6 groups). The graduates were distributed in different organizations.
You need to learn both, but remember that only R has a reliable and proven gateway to MQL.
Go to the next level - keras/tensorflow/. There are so many ideas, I wish I had enough knowledge and time to master them.
Good luck
A reliable MQL gateway is not a problem with anything at all. There is a problem here, but it is common for all MQL gateways.
Basically I already have a good command of both R and Python. If I want to get acquainted with the module-packages, this is where it is even worse.
The package-modules themselves are not the ideas, but only the apparatus for implementing the ideas. For ideas, a knowledge of principles is sufficient.
It is bad when there are no ideas at all, no ideas at all. But it often happens that way when you finish some work but haven't yet started a new one.
Check out the packagevarbvs . The package implements fast algorithms for fitting Bayesian variable selection models and computing Bayes coefficients, in which the outcome (or response variable) is modeled using linear or logistic regression. The algorithms are based on the variational approximations described in" Scalable variational inference for Bayesian variable selection in regression, and its accuracy in genetic association studies" ("Scalable variational inference for Bayesian variable selection in regression, and its accuracy in genetic association studies" by P. Carbonetto and M. Stephens, Bayesian Analysis 7, 2012, pages 73-108). This software has been applied to large datasets with over a million variables and thousands of samples.
It selects predictors well and builds good models.
Good luck
Thank you! I already have it in my piggy bank. I like the speed - only 2 seconds (by comparison saget-rfe takes 16 minutes).
Hare COPY!!!! It's time to act.....
They also advise to pay attention to the loss function for regression problems
A fresh book on deep learning is out in Russian:
Goodfellow Y., Bengio I., Courville A.There is another book with the same title in Russian on ozone - https://www.ozon.ru/context/detail/id/142987816/
There is one more book in Russian with the same name in Ozone - https://www.ozon.ru/context/detail/id/142987816/
Thank you. I purchased it earlier. It's a handbook.
I advise everyone to work through it.
Good luck
the pattern changes chaotically and the deviations in the patterns grow exponentially with time
any approximator (except, partly, RNN or LSTM) is unable to solve such problems
All articles on statistics, with attempts to apply them to the market in their current form - can be thrown out and do not pay any attention to them
the main efforts should focus on methods of working in a non-stationary environment, one of which is suggested by Alexander (assuming you don't have features that are stationary affecting the quotient, which cannot be extracted from the quotient itself, a-priori)Bravo. Understanding the essence of the problem brings you to a new level.
Here's an idea - maybe the solution lies in fairly simple things such as fundamental analysis and the current price position in relation to the historical min and max? First of all, the price of the base currency is influenced by news factors, it is difficult to put them into a code, I do not know if there are such advisors on the news? If there are, then most likely they carry out fund. analysis by regular reports of central banks of the countries, whose currency in the pair is the BASIC, in principle there is a small list of indicators needed to evaluate the BASE of the base currency: here we get 1 parameter - the motivated weakening or strengthening of the base currency according to fund. analysis. Similarly, we study the fund. analysis of the CONTROLLED currency. The final changes e.g. through the ratio of the changes in the BASE of each currency in the pair according to the Fund. analysis should indicate in favor of this or that BASE of the currency in the pair and in this way the signal is formed. According to this analysis, large financial institutions redistribute currency risks, buy or sell currencies of the country whose economy is weakening according to fund. analysis. It makes sense. All this applies to a long-term strategy.
The second indicator is the price position of the currency pair in the here and now. If we carry out gradation by horizontal lines, then you can set some weight to each of these lines for the purchase and sale, and here I see a more appropriate tool for medium-term trading.
And the third parameter is of course the indicator. This is a quick signal. But it does not give any predictions, as you have correctly summarized in the previous 854 pages of this interesting topic.
The task - how to relate a long term signal - BASO, medium term - the weight of the horizontal line near which the price is here and now (let it be Fibonacci lines for example) and the third parameter - the signal from the indicator.
These very criteria, in my opinion, are the most important and can really teach the NS trading system. The only difficulty is that you need a team for that - so make friends-friends among the financiers-fundamentalists or macroeconomists, who will help you choose the right algorithm for processing their reports data flow and appropriately interpret them, with respect to your NS trading system, by the way for the analysis of stock price behavior you will also need a simple economist or financier - these are specialists in economic subjects. The subjects of the economy: the state, legal entities and individuals.
To teach by the history of quotes - well, you have successfully passed this way, the experience is gained. Now we understand that we have to try to suggest to the system the algorithm of data collection to get at least these three basic parameters, in the analysis of which we can look conditionally into tomorrow (make an assumption, set the weights of events development (forecast) for deals of different timeframe) and according to these parameters your NS system will be able to make a motivated decision to enter in buying or selling, including the type of deal - fast, medium, long - and the nature is set by a simple take profit level, volume or multiplier.
Something like this... It's complicated, but I guess you're not looking for easy ways)))
The calendar API seems to have been announced, but it's not in MT5 yet
It would be interesting to use a news feed... I'm not sure if the results would be satisfactory, but just for the sake of curiosity
+ you need to google intensively for new research on non-stationarity. In RL this research is being intensively done at the moment, i.e. the topic is still evolving, so I'm sitting on it for now. The simplest example is effective various feedbacks, analytically they cannot be calculated or imagined, so only through multiple experiments :)