Machine learning in trading: theory, models, practice and algo-trading - page 1835

[Mihail Marchukajtes]

Valeriy Yastremskiy:

It seems that the grail topic there has not died down, there people are cracking))))

the grail is just a grail to strive for it BUT you will never reach it.... eternal topic....

[mytarmailS]

Valeriy Yastremskiy:

in the understanding that it is a division of a set into subsets. As here apparently not.

in the idea that the pink section should be divided into 10 parts apparently)

Not into 10 parts but into 10 subsets, subsets can have any number of parts, well clustering, normal clustering

[Mihail Marchukajtes]

I know. Asked the administration of the KC service about the possibility of creating a private server purely for the local residents. If anyone does not agree or wants to aggressively express their dissatisfaction with his opponent invites him to the map and fight at knifes. ***there is no place.

Well, let's wait for a response from the administration, what they say...

[Valeriy Yastremskiy]

mytarmailS:

not into 10 parts, but into 10 subsets, subsets can have as many parts as you want, well clustering, normal clustering

Subsets are clusters - paternets - repeating chains of events. That is you mark a corridor in the sliding window at every last price and look for events in that corridor, then you look for the chains of events - paternoster clusters among them. And then you check these patterns for profitability.

Right?

[Mihail Marchukajtes]

Valeriy Yastremskiy:

Subsets are clusters - paternas - repeating chains of events. I.e. at each last price in the sliding window you select a corridor and in this corridor you look for events, then you look for chains of events - paternas - clusters among them. And then you check these patterns for profitability.

Is that so?

No. Patterns are specific conditions that have formed. Clusters are signs for grouping. Patterns and clusters are not the same thing, unless it's a special case.... Where the number of clusters is equal to the number of patterns and the grouping is done by belonging to one or another loss-cluster. Otherwise, they are different things...

[mytarmailS]

Valeriy Yastremskiy:

Subsets are clusters - paternas - repeating chains of events. I.e. at each last price in the sliding window you select a corridor and in this corridor you look for events, then you look for chains of events - paternas - clusters among them. And then you check these patterns for profitability.

Is that so?

Yes, exactly!

[Valeriy Yastremskiy]

mytarmailS:

exactly!

On the one hand you narrow the search area, on the other hand you lose results at the ends, and if the area is too wide the search will be long.

Search/identify clusters to find patterns first in what part of the row.

In general for a stationary area at first glance should work, but when you change stationary conditions it should stop. or you need to have patterns in advance for other stationary conditions.

similarity problem. This is when the beginning of the pattern is the same, and they differ only in the last event.

And there remains the issue of recognizing the pattern before it ends (by incomplete features) in the test/real.

[Rorschach]

What does 1 bar mean for Ma(100)? But such a shift is enough to make the dumbest system work (although logically it should be at least 50). Careful with bikes and cycles.

[Maxim Dmitrievsky]

https://www.mql5.com/en/forum/321503

on the topic of developing examples of differentiation from my article, liked

I found it by accident through google, I wouldn't have known.

[Valeriy Yastremskiy]

Maxim Dmitrievsky:

https://www.mql5.com/en/forum/321503

On the topic of developing examples of differentiation from my article, enjoyed

I found it by accident through Google, otherwise I would have never known.

New possibilities give new ideas / thoughts))))