Machine learning in trading: theory, models, practice and algo-trading - page 689
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 want to rush things, but it seems to me that models built on inputs with maximum output information work much longer. It used to be difficult to meet the 2 week threshold, but now this is the result of the last 2 weeks.
Quite decent, I think. But more tests are needed....
One more interesting lib on plusses with RL. If you have any interest in rewriting it in mql, let me know, I am still studying the theoretical part
http://mlpack.org/gsocblog/deep-reinforcement-learning-methods-summary.html
In the rigid formulation of the law of causality that says, "If we know the present precisely, we can calculate the future," it is not the second part but the premise that is false. We fundamentally cannot know the present in all its details.
At one time I misjudged the idea of principal components: I thought that they should lead to improved prediction accuracy.
But then I came across several articles which claim that the accuracy remains the same, but the predictive ability of the model stabilizes (model overtraining falls). Especially breakdown into main components in relation to target variable.
Yeah, yeah... I guessed that already... Thank you all in general, I'm getting pretty damn good at R..... The important thing is to understand the syntax of....
It's not the syntax.
Unlike C-like languages, R is a vector language which has no concept of scalar. If one always keeps this in mind and can write a <- b, where b is any object, in the simplest case a vector or a matrix, and loops are not needed for assignment, it turns out that one can write very short and information-intensive programs.
There is one more trick.
Suppose there is a vector that contains 5 numbers. Each of the numbers takes its place in the sequence 1, 2,... 5. These places. And these places can have names other than place numbers. These names are also collected into an independent vector, on which you can also do operations.
Very interesting are vectors where the names of places of values in the vector are timestamps. It's a separate type of data - time series.
Yeah, yeah... I guessed that already... Thank you all in general, I'm getting pretty damn good at R..... The important thing is to understand the syntax of....
https://msperlin.github.io/pafdR/importing.html
To make it clear what I was talking about yesterday.
Look at the bar chart of the quote flow intensity for the pair AUDCAD in the sliding window = 4 hours:
How can you trade with that kind of bullshit on the inputs?
I repeat once again - you have to do your best to transform the input stream to bring the intensity to a Poisson distribution.
Without this key point, all efforts are doomed to failure.
And I know how to do it.
Moreover, after such transformation of intensity, histograms of increments take strict form and there is no need to logarithm them, etc.
I.e. having solved the problem with the quote flow, you can easily work with the purest increments, which is what you need.
That's it!
To make it clear what I was talking about yesterday.
Look at the bar chart of the quote flow intensity for the pair AUDCAD in the sliding window = 4 hours:
How can you trade with that kind of bullshit on the inputs?
I repeat once again - you have to do your best to transform the input stream to bring the intensity to a Poisson distribution.
Without this key point, all efforts are doomed to failure.
And I know how to do it.
Moreover, after such transformation of intensity, histograms of increments take strict form and there is no need to logarithm them, etc.
I.e. having solved the problem with the quote flow, you can easily work with the purest increments, which is what you need.
That's it!
This is not pussyfooting around. Where is the example after the conversion? At least a picture of what you got there. And it is desirable that the picture shows the balance of funds. Because this is the ultimate goal of any market work, no matter how big or secret it would be....
And here's a question that I have, well, not so easy. There is a table, 10 by 10. How to get a table without the first column or the fifth. That is, you need to get the object without the i-th column????
And here's a question that I have, well, not so easy. There is a table, 10 by 10. How to get a table without the first column or the fifth. That is, you need to get an object without the i-th column????
m[,-5]
m[,-5]
Yes, yes thanks....