Machine learning in trading: theory, models, practice and algo-trading - page 1994

 
Maxim Dmitrievsky:
There are no layers there, it's a tree booster

Man, the function is not linear and complex between the source and the forecast. But initially the goal is to match and correct forecast. that's why the correlation is usually positive. Not linear. And it makes no difference what the prediction algorithm is. The data is prepared for the purpose of the correct prediction. But all this is not yet enough for a reliable forecast.))))

 
Valeriy Yastremskiy:

Man, the function is not linear and complex between the source and the forecast. But the initial goal is to match and correct prediction. so usually the correlation is positive. Not linear. And it makes no difference what the prediction algorithm is. The data is prepared for the purpose of the correct prediction. But all this is not yet enough for a reliable forecast. ))))

F-y simple linear, p squared errors. As Elibrarius pointed out - let's just call it a coefficient. But to the person who asked this might be incomprehensible, so I converted it to a percentage. We can continue to discuss this very important topic for everyone 🤣🤣🤣🤣
 
Maxim Dmitrievsky:
F-square simple linear, p-square error

Well we are talking about different things. p-squared errors is an estimate of the forecast. not the dependence - a function of the raw data to the forecast. in the estimate is simple, the data are real and forecast, but the dependence on the raw forecast data is complicated. And it usually isn't made. It's complicated. It's easier to estimate the result.

 
Valeriy Yastremskiy:

Well, we're talking about different things. p-squared error is an estimate of the forecast. not the dependence - a function of the raw data to the forecast. in the estimate is simple, the data are real and forecast, but the dependence on the raw data forecast is complex. And it usually isn't made. It's complicated. It's easier to estimate the result.

Well he asked the result, the model will show the dependence or not. The classic one does not. Maybe some sophisticated model will.
 
Maxim Dmitrievsky:
Well he asked for the result, whether the model shows dependence or not. The classic one does not. Maybe some kind of sophisticated model will.

Man, I'm getting a little uncomfortable... I didn't have to lecture you.)))) Let me ask you a question. Correctly understood the python tutorial. The global locality of names is determined by the place of the name declaration, but the actions with them if in someone else's namespace are determined by global locale prefixes. and if without prefixes, the namespace name is prioritized. Interesting namespace logic, especially after BASIC))))

 
Maxim Dmitrievsky:

on five previous values:

60% accuracy

on 10 preceding deteriorates, accuracy 50

accuracy 57 on the 3rd.

Blue initial, orange prediction. The last 300 values.



Thanks Maxim!

Actually, I forgot to say, it is enough that the model predicts the next sign (+-) of the series, target + plus value. And the specific accuracy is not important.

A sample of length n = 54 , covers all possible values.

 
Valeriy Yastremskiy:

Man, I already feel a little uncomfortable... I was kind of not lecturing)))) Let me ask you a question. Correctly understood the python tutorial. The global locality of names is determined by the place of the name declaration, but the actions with them if in someone else's namespace are determined by global locale prefixes. and if without prefixes, the namespace name is prioritized. Interesting namespace logic, especially after BASIC))))

Yes
 
Evgeniy Chumakov:


Thank you Maxim!

Actually, I forgot to say, it is enough that the model predicted the next sign (+-) of the series, target + plus value. And the specific accuracy is not important.

The sample length n = 54 , covers all possible values.

I will do it later today, if I have time.

 
Maxim Dmitrievsky:
Yes.

Thanks. It's worth something. The explicit freedom in variable types turns out to be an explicit type indication. Although if there are no identical names, it should work without prefixes.

 
Valeriy Yastremskiy:

cps. everything is worth something. explicit freedom in variable types turns out to be an explicit type indication. Although if there are no identical names, it should work without prefixes.

Using a large number of globals is not good practice, and there is no point. It's necessary to see the programming patterns. The explicit type indication has no effect, it can still be changed at any time