Machine learning in trading: theory, models, practice and algo-trading - page 1726

[mytarmailS]

Maxim Dmitrievsky:
the question is whether it's possible to trade every minute separately ) whether there's a profit more than the spread

Well, yes, that's the question of course ))

Well, experiment is the criterion of truth.

[Mihail Marchukajtes]

Rorschach:

It takes a system dangling at zero. It turns out that equity serves as an oscillator. When it deflects to a certain level, it enters with real money.

Well, then Z-score is of great help to you. I was just counting it today :-)

[Maxim Dmitrievsky]

mytarmailS:

Well, yes, that's the question of course ))

Well, experiment is the criterion of truth.

Yeah, it's a cool topic, but if you collect returns not by time, but by a specific value of some indicator, the same round levels, as Rorschach suggested, or some other condition... it's a mining farm

another binge

[mytarmailS]

Maxim Dmitrievsky:

Yes, it's a cool topic, but if you collect returnees not by time, but by a specific value of some indicator, the same round levels, as Rorschach suggested, or other conditions... it's a mining farm

just another bogeyman

Well it turns out the usual decisive rule, which builds for example Forest, no?

I say, you can even choose any particular rule and collect dataset for it, and teach the model suto for this rule, I have here

#16577 for example, I did just that, +- 800 models in one robot... But the result is also rubbish

[Maxim Dmitrievsky]

mytarmailS:

Well it turns out the usual decisive rule, from which for example forrest is built, doesn't it?

I say more, you can even choose a specific rule and build dataset for it and teach the model exactly for this rule.

#16577 for example, I did just that, +- 800 models in one robot... But the result is also rubbish

Well, sort of an overrun of conditions, yeah. I don't know what's different, it's just an original method. A new way to optimize :)

[Rorschach]

On the subject of spectra and recognition.

[Mihail Marchukajtes]

By the way Max. I saw you posted the cointegration indicator the other day, I confess that I did not look at it yet, but in general I was interested in the topic back in NS. So, there was an instrument in NS that calculated cointegration and built Lysajo figures, those notorious lines on the oscilloscope screen. The task was to draw a circle with these lines. The flatter the circle, the steeper it was. It was interpolated in the way that the frequency found was ahead of the quote and remained so for some time. Literally by two or three inflections. But unfortunately it was impossible to optimize it and I had to do it manually. One time I happened to obtain such a figure (circle) and indeed the parameters found for the filter were ahead of the prices for some time. I had to do it manually and could not get such figure any more, so I gave it up. But the way the frequency found at that moment worked was a miracle. I already told you about it. Maybe it would be worth doing some work in this direction. What do you think?

Yeah.... Max, I am not even lazy and found a screenshot of this window, where it was necessary to pick up the load. But the hands to do this is very difficult because one of the parameters is the window for analysis, of course listened to the bar and find the right window was just not realistic.

What do you think?

[Mihail Marchukajtes]

As you can see the result of the selection is the found chatsota which goes to the price preemptions.

[Maxim Dmitrievsky]

Mihail Marchukajtes:

What do you think?

I don't understand anything, I'm going to bed.

[Mihail Marchukajtes]

Maxim Dmitrievsky:
I don't get it, I'm going to bed.

Listen, I think we already discussed this in '18. I just read it in the indicator thread. So, using this tool we find parameters for CSSA in relation to quote frequency and in this way we tune filter in such a way that it saves its oscillations values in relation to quotes oscillations for a certain short period of time.