Discussion of article "Statistical Distributions in MQL5 - taking the best of R" - page 3
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
There is an obvious picking on the wording "analogue".
In the article it is analogue, and a complete analogue at that. In R, almost everything goes through vectors. These are issues of concise syntax, for which, in particular, R is loved so much and deservedly.
And it has nothing to do with the article. It's nagging in its purest form.
Again, step aside. I'm not picking on you, and you have NO idea what I'm writing about.
I have a question to the author or Renat, with whom we discussed the use of R in the thread to which he gave a link.
It is claimed that this is an analogue of the R function, which is specified in the text.
What is the result of calling the specified function in MQL? A scalar? A vector?
In this case, the function works on one value and returns the result of processing this value. To process an array, you need to run through its elements in a loop.
Tomorrow we will check everything and show an example in MQL5. Unfortunately, I don't have R at hand to check it.
Of course, array processing should be added to all similar functions, which we will do.
Thank you for checking and paying attention to the lack of vector operations.
In this case, the function works on one value and returns the result of processing this value. To process an array, you need to run through its elements in a loop.
Tomorrow we will check everything and show an example in MQL5. Unfortunately, I don't have R at hand to check it.
Of course, array processing should be added to all similar functions, which we will do.
Thank you for checking and paying attention to the lack of vector operations.
Dear Renat!
Vector is a trifle, and with trifles I wouldn't post here.
The matter is much more complicated and that R has extremely significant differences from µl.
On the surface are what you called: vectors. There is no concept of a scalar in R, but vector and matrix operations. But that's on the surface. And it's not the most important thing.
The matter is much more serious. Namely: in the concept of "object", which is available in R. This notion is taken to its maximum: everything can be an object: data, scripts, functions that are in the workspace, as well as the configuration of the computer on which it is executed.
I understand perfectly well that if you start with statistics, you cannot do without the given functions - they will not be respected, they just have to be.
But if you take the caret package containing a hundred and fifty models related to trading and see WHAT the functions listed there return.....
So it's not about vectors. The point is the principal possibility to repeat what is available in R by means of MKL. Take a look at the structure of objects in this package. I think that after that you will understand the sense of my suggestion to rewrite the user part of the terminal in R and under R. It seems to me that it will not be too costly, as R is very friendly with C.
As soon as you put a package called MT-R in CRAN, docked with response parts from brokers, you will stop being "the first guy in the village and become the second guy in Rome".
Dear Renat!
Vector is a minutiae, and with minutiae I would not post here.
The matter is much more complicated and that R has extremely significant differences from µl.
On the surface are what you called: vectors. There is no concept of a scalar in R, but vector and matrix operations. But that's on the surface. And that's not the most important thing.
You think I'm not aware of this?
We have a completely different class of language (classical, strictly typed) and we won't create multifunctional dynamic objects in it. But the corresponding functionality of scalar and vector operations in the standard library will certainly be there.
The set of mathematical functions should be sufficient for writing target applications. Not for research (this is done in R and similar systems), but for writing target working programmes.
Nobody is going to make a copy of the dynamic language R.
I would like to make a reservation - I am not criticising the idea itself. the more opportunities, the more likely the community that uses it.
I am far from machine learning, from neural networks. I am, for example, more familiar with pair trading. So, for this purpose, we can use linear regression (available in aglib), principal component method (available), orthogonal regression (not found), kalman filter (not found). Then after building a model it should be estimated somehow - adf test, Engle-Greiners, Johansen. there is nothing like that. one of the most popular books on this method is by Ernest Chan (the book contains all examples on matlab), but there is already a port on R of all examples.
Let's say there are options (let's hope that they will be publicly available someday). there are methods of options marketmaking. there are already ready solutions in R for this, at least for a part of such a problem. https://www.youtube.com/watch?v=8jJNZAMXWic perhaps some traders will be interested in such a task.
What is the point of all the above? The point is that there are a huge number of specific tasks that cannot be covered, i.e. the user will have to either port ready-made solutions to the µl environment himself or wait until it is done by the developers, which is very inconvenient and unproductive. It's like, say, suggesting to port new libraries to the java environment. you will answer - why? we have a ready solution in our environment! and that's right. the same with users - who will be ready to port their already working solutions to another environment?
This is a bit rambling, and maybe I didn't write everything I wanted to write, but that's how it is.
I would like to make a reservation - I am not criticising the idea itself. the more opportunities, the more likely the community that uses it.
I am far from machine learning, from neural networks. I am, for example, more familiar with pair trading. therefore, it is possible to use linear regression (available in aglib), principal component method (available), orthogonal regression (not found), kalman filter (not found). then, after building a model, it should be evaluated somehow - adf test, Engle-Greiners, Johansen.
What the fuck?! The goals of the mat. bible are clear.
Forum on trading, automated trading systems and testing trading strategies
Discussion of the article "Statistical Distributions in MQL5 - Taking the Best of R".
Renat Fatkhullin, 2016.10.09 16:26
The set of mathematical functions should be sufficient for writing target applications. Not for research (this is done in R and similar systems), but exactly for writing target working programmes.
What the fuck?! The purpose of the maths bible is clear.
You start by answering simple questions - do YOU personally use anything ? Are you going to use it in the future? Or do you just want to insert YOUR 5 kopecks?)))
Why am I asking this? Because according to Renat this is just the beginning and the developers will not limit themselves to these libraries.
You start by answering simple questions - do YOU personally use anything ? Are you going to use it in the future? Or do you just want to insert YOUR 5 kopecks?))))
Forum on trading, automated trading systems and testing trading strategies.
Do you use CExpert when creating robots?
fxsaber, 2016.10.01 16:15
I can imagine. There is an excellent expert who has eaten a dog in pattern recognition, Big Data, Machine Learning and the rest.
But he has never encountered the financial market. It just so happens. Super-expert in mathematical languages, training above all praise.
And suddenly he finds out about the financial markets. "That's it, I'm going to beat them all, with my baggage and experience. With my mathematical models and knowledge of mathematical languages."
И ... poof! What does all this rubbish, with all due respect, have to do with the creation of robust TCs?!
Some people think they didn't create robust TCs because they didn't have enough knowledge. I'll study R, and then I'll definitely create one! Well studied, well studied, well spun price series and so what?
And the result is the same, that you know or don't know R. These are financial markets, not pattern recognition.
My vision of this situation is simple. Renat ordered - they started to do it. MT5-R gasket developers will not make - everything is clear here. Anyone has the ability to make such a gasket himself? - Yes, it's just as clear here. Resources for the mat. bibla will be taken away or a new Quantum person will be hired for this case - I don't know. Most likely, the staff has expanded and human resources to eliminate bugs have not decreased.
Will the whiners, haters and beggars disappear somewhere? -No, never. Do you want to call them weirdos? -yes, of course. Can/should I write about them? -only if they're annoying and there's no other way to blow off steam.
Looking from the subjective viewpoint of a self-taught programmer who knows only MQL really well, I would like to note a strange "skew" in the trends of the algo-trading community.
The essence of the view is as follows: local robot developers choose "increasing" computing power, attaching neural complexes and using higher mathematics to analyse market processes as a priority in developing the capabilities of Expert Advisors.
At the same time, developers are moving further and further away from classical technical analysis, which does not require such tools at all. Perhaps, this is because classical technical analysis is designed for a different pace of trading, for other strategies and for other player deposits....
Small players are forced to play very fast, without building long-term strategies, they close stops and therefore their trading is constantly moving towards acceleration. Their environment is a very fast market in which the processes are much less predictable than in the slow market - the market of big players. Of course, trade automation and the mathematical basis used in it are their "trump cards" in the battle with the big players for profit, but by their actions they themselves create the chaos in which, in the end, the invested computing power sinks, the higher mathematics fades, and automation provides a game of "roulette".
The lack of a rational basis for decision making in fast trading renders meaningless what should give players an advantage - automation.
The inevitable plunge into the element of randomness when trading faster ultimately negates the point of automation, and turns maths into a useless application of the mind. The elements cannot be predicted by any means, if there is not at least a hint of some order hidden in it, but this "hint" is created by ourselves, making well-considered decisions in trading. All decisions made on the basis of abstract formulas will multiply chaos on the market.
I propose to return to the origins and remember what moves the market. First of all, the market is moved by decisions based on public re-cognition of the value of things, not abstract mathematical calculations of future price movements.
If this is forgotten, the market will die.
I don't understand something
San Sanych, R is here for publicity and nothing more. MT extends mat functions and nothing more. I don't apply it, in itself, in isolation from the rest. Probably good (extending mat-functions is always good), but I personally don't give a shit about them, in isolation from the rest of the computational capabilities of R and SciLab.