Machine learning in trading: theory, models, practice and algo-trading - page 1718

[Maxim Dmitrievsky]

mytarmailS:

I'll tell you a secret))

they're all steady ))))

except the point is )) the point is different ))

but you wouldn't understand it because you need to broaden your horizons ....

And you're looking at everything from a position of know-it-all and therefore closed to the new...

Take a couple of components, check the variance and the average on both OS and OOS, you can break it down by hours, etc., and report back

I'll spit out a finished bot for you right away, after you give me the diffur for the component

[mytarmailS]

Maxim Dmitrievsky:
I read your Kant and Hegel. Show me the cycles in these rows and that they are stored in the new data.

There are no cycles!!! You made them up yourself, moreover I wrote that there are no cycles, there are cycles in volatility but they are useless, also I wrote that your econometric decompositions are rubbish because they imply seasonality (cycles), and I use what is suitable for nonstationary series (those without cycles)

And you're all cycles, cycles ... do you drink there?

Read what spectrum analysis is, what it's used for and how it's used, moreover I wrote what you can do with it and there was no talk about cycles, and you're still talking about cycles, cycles ...

I said you've got the know-it-all syndrome, you've read something, you've got a stamp in your head and you're like a bio robot, you don't hear new information, even if it's shouted in your ear.

[Maxim Dmitrievsky]

mytarmailS:

There are no cycles!!! You made them up yourself, moreover I wrote that there are no cycles, there are cycles in volatility but they are useless, also I wrote that your econometric decompositions are rubbish because they imply seasonality (cycles), and I use what is suitable for nonstationary series (those without cycles)

And you're all cycles, cycles ... do you drink there?

Read what spectrum analysis is, what it's used for and how it's used, moreover I wrote what you can do with it and there was no talk about cycles, and you're still talking about cycles, cycles ...

I say you have the know-it-all syndrome, you've read something, you've got a stamp in your head and you're like a bio robot, you just don't hear new information, even if it's shouted in your ear.

TC can only be built on cycles, that is, on what is repeated. It's out of the question.

maybe you have the wrong idea about cycles.

Examples of singling out cycles on SB and on price increments

https://www.mql5.com/ru/forum/286022/page169#comment_15898101

https://www.mql5.com/ru/forum/286022/page169#comment_15898212

[mytarmailS]

Maxim Dmitrievsky:

Maybe you have the wrong idea about cycles.

Maybe )

[Alexander_K2]

Maxim Dmitrievsky:

TC can only be built on cycles, i.e., on what is repeated. That's out of the question.

Maybe you have the wrong idea about cycles.

Examples of the cycles in the SB and in price increments

https://www.mql5.com/ru/forum/286022/page169#comment_15898101

https://www.mql5.com/ru/forum/286022/page169#comment_15898212

Indeed it is.

[mytarmailS]

Maxim Dmitrievsky:

TC can only be built on cycles, i.e., on what is repeated. That's out of the question.

Maybe you have the wrong idea about cycles.

Examples of the cycles in the SB and in price increments

https://www.mql5.com/ru/forum/286022/page169#comment_15898101

https://www.mql5.com/ru/forum/286022/page169#comment_15898212

I don't understand anything (((

If I look at the statistics the real number of forecasts is higher, the stronger they are, the better their quality, the better their forecasting.

So what did you want to tell me with these links, I do not understand (

i have a "real" target, the direction of the zigzag, but not the increment from the increment

classification

training 300 candles, 90 chips

normal forrest on 100 trees

training

    OOB estimate of  error rate: 1%
Confusion matrix:
    -1   1 class.error
-1 107   2 0.018348624
1    1 191 0.005208333

new data

 PRED
     -1  1
  -1 10  0

10 out of 10 ))

I cannot train much and cannot predict much, because the model dies very quickly, ideally I need to retrain on every bar

Such an adaptive filter turned out with one step prediction ))

[Forester]

10 hits on the OOS are statistically insignificant.
300-1000 repetitions can tell you something.

Do another 10 training and OOS on completely different data. Get a walking-forward analogy.

[mytarmailS]

elibrarius:

10 hits on OOS is statistically insignificant.
300-1000 repetitions can tell you something.

Do 10 more training and OOS on completely different data. Get a walking-forward analogy.

Yes, yes, I understand... I do.

[Maxim Dmitrievsky]

mytarmailS:

I do not understand anything (((

Yes, you can predict the forecasted series, the stronger you did, the better the quality, you can predict that the glazing, I dabbled in it myself, but when you return the forecast to its original, I want to cry))

So what did you want to tell me with these links, I do not understand (.

What are we talking about? Decomposition into components is differentiation, so to speak. If you can predict something that is larger than the spread, that's already a profit.

Tell me a digital filter that its difference with the price forms some stable cycles, which last for a long time. I don't understand it, how do you look for them?

[Maxim Dmitrievsky]

Let's dig into how this is implemented... here's a link, for example

https://medium.com/@sadatnazrul/digital-signal-processing-for-predicting-stock-prices-4be247a09514