Machine learning in trading: theory, models, practice and algo-trading - page 782
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
Spit on the mkl - not enough a huge number of the most different tools, you're going to stumble all the time in different nonsense like acf and a lot of other things. mkl later, when you have a working model, and this may not be able to get at all. In R you just know that it's not working, in MQL you don't, because you lack the tools.
In MQL you need to run it on history, in R it will be hard to do. Again, the trades can be seen visually. And the combination of R and MQL will probably be very slow in the optimizer?
In MQL you need to run it on history, in R it would be hard to do. Again, trades can be seen visually. And the combination of R and MQL will probably be very slow in the optimizer?
It is a long way to the run. First, we have to select the variant, and there are plenty of variants, and the most important thing here is the speed of idea implementation. Accuracy, having a tool at hand...
There is another nuance: in R you can test the model in a way that is impossible in µl. These tests will give a theoretical justification for the future behavior of the TC. μl at the end.
PS.
The graphics in R are much richer, you can use it instead of µl, it's just not necessary, not that stage. But to quickly see how any vector or their superimposed set looks like is a sneeze and very useful.
To each his own and there is no need to confuse them.
It's a long way to run. For the beginning it is necessary to choose a variant, and there are extremely many of them, and here the main thing is the speed of realization of the idea. Accuracy, availability of the tool at hand...
There is one more nuance: in R it is possible to test the model as it is impossible in μl. These tests will give a theoretical justification for the future behavior of the TC. μl at the end.
PS.
The graphics in R are much richer, you can use it instead of µl, it's just not necessary, not that stage. But to quickly see how any vector or their superimposed set looks like is a sneeze and very useful.
To each his own and there is no need to confuse them.
Considering the above. I have been wondering for a long time if it is possible to see in R, over the whole history, how the forecast of the same ARIMA has come true, to go through the best periods for it,
that is, pick the best settings, like MQL? Or at least to dump this data into a file.
I am not asking anyone to write anything for me, but it seems to me that I am an excellent algorithmologist. I remember one of the lecturers who said that in companies that create software algorithms are valued more, because they invent, while programmers only implement. As a rule, if a person is engaged in it, he combines these two qualities. I fundamentally can't and don't want to. But I am a good programmer in terms of ideas, of which 99% then sifted out. But you need only one idea. :-)
Now the main thing is to decide on the utility. It should be simple, but as flexible as possible. A large number of features and functions.
You won't believe, all my life I've been working only with classification, and I got bored all of a sudden. I had to redo everything three times during the night, because the error in the calculation of the deviation (htcgtrnelibrarius) has entailed there and more, but already now it is clear that the sleepless night was not in vain. And now it has become kind of boring, and it would be interesting to try something new.
Regression. There are, of course, differences both in the approach and in the possibilities of the information obtained, BUT the principles of one method can be interpreted into the other. Not to make gross and hidden mistakes.
Therefore I wait when you will be defined, because only I see all new packages, some better and some worse. Again, we need the most flexible.
Doc, Sanych, Trickster, too, let's call. A kind of combat unit. I move the method you implement it. But we need Assassins. Here I bow my head to such people. Sometimes the quality of their decisions is surprising. For the first time I encountered such things in Reshetov's code, when I was still at the beginning of understanding models and how they work. There was a part of the code, which I couldn't understand at first glance, and when I scrolled it through my head step by step, I was amazed at how he did it. I would have written it three times as big and it would have been glitchy, but it was so simple and to the point. There was such a trader klot back in the day. He once said he could "program the hell out of him" and he was really good. A really good programmer. But it was long ago.
In general, let me know if you're ready to work as a team and let's get started. It will be interesting. Classifik solves the problem of regression. :-) CLASSIFIK HA HA HA. That's hilarious. what did i call myself.... Almost busted my belly!!!!!
Given the above. I've been wondering for a long time if it is possible to see in R, on the whole history, how the forecast of the same ARIMA has come true, to look through the best periods for it,
that is, pick the best settings, like MQL? Or at least reset those data into a file.
It is very inconvenient to test regression models in MQL by standard means, and ARIMA is just a regression model.
Classification models give an answer in the form of several classes - e.g.
"-1" and "1", or
"0" и "1",
"buy" and "sell",
"long" and "short",
etc.
All of this can easily be added to the trading logic, see the profit chart, and evaluate it with a Sharpe Ratio for example.
Regression models do not give the trade direction, but the forecast of the price itself. And the closer the forecast is to the real price, the better. Of course, in MQL EAs, it is possible to trade on the principle of "if the forecast is higher than the current price, then I buy. If the forecast is lower, then I sell. But it will completely miss an even more important metric: how close is the forecast to the current price? There can be two models that give the same trade direction, but the second one will give answers closer to the real price. In R, you can see it immediately through the estimation of R2 and take the correct model. And in the mql tester they will be identical for you, that's bad.
> Is it possible to see in R, over the entire history, how a forecast of the same ARIMA has come true, to look through the best periods for it?
Yes, for example, by saving bar history from mt5 to csv file, import it into R, and there sliding window train on some period and test it on the next one, and so do cyclically shifting this training window.
I move the method you implement it.
Thanks for the offer, but I will decline. I also have a list of ideas for decades to come.
It's very inconvenient to test regression models in MQL by standard means, and ARIMA is just a regression model.
Classification models give the answer in the form of several classes, e.g.
"-1" and "1", or
"0" и "1",
"buy" and "sell",
"long" and "short",
etc.
All this can easily be added to the trading logic, see the profit chart, and evaluate it with a Sharpe Ratio for example.
Regression models do not give the trade direction, but a forecast of the price itself. And the closer the forecast is to the real price, the better. Of course, in MQL EAs, it is possible to trade on the principle of "if the forecast is higher than the current price, then I buy. If the forecast is lower, then I sell. But it will completely miss an even more important metric: how close is the forecast to the current price? There can be two models that give the same trade direction, but the second one will give answers closer to the real price. In R, you can see it immediately through the estimation of R2 and take the correct model. And in the mql tester they will be identical for you, that's bad.
> Is it possible to see in R, over the entire history, how a forecast of the same ARIMA has come true, to look through the best periods for it?
Yes, for example, by saving bar history from mt5 to a csv file, import it into R, and then use a sliding window to train on one section and test it on the next one, and then do this cyclically shifting the training window.
That's right Doc. Regression, in addition to the direction of movement, also gives the degree of that movement, as opposed to classification, which only gives the direction. Isn't it a bit cheeky of regression to know the future, which is not certain???? Classification doesn't allow for that.....
So what do you think of my suggestion above?
Publicly, right in this thread. Step by step. And I will act as an imprisario and I will do this ballet according to your recommendations, too. So to say, I will help my friend and myself :-)
Thanks for the offer, but I will decline. I, too, have a list of ideas for decades to come.
You didn't get the gimmick of a classification engineer's experience in the regression direction......
From me ideas at the level of algorithms that you will apply to your models. Your TCs and directions. But with one condition!!!!! Publication of results at each stage.....
Absolutely everyone is welcome, but only Experienced programmers without stupid questions what is the spread!!!! :-) Op.... funny again :-)
Regression. There are differences, of course, both in the approach and in the capabilities of the information obtained, BUT the principles of one method can be interpreted into the other. Not to make gross and hidden mistakes.
Therefore I wait when you will be defined, because only I see all new packages, some better and some worse. Again, we need to be as flexible as possible.
If I'm not mistaken, the main problem is that no one has ever returned from those who went to the "packages" (this is somehow ambiguous).
You'll have nothing until you have a robust idea... all you've been doing all these years - nothing.
because you've been running it with your supposedly good features, but it's not your features that are good
the regularities exist in the real world, not in the world of bags, so look for them, and if you find them, you can even code them yourself
or better yet, forget it