Is there a pattern to the chaos? Let's try to find it! Machine learning on the example of a specific sample. - page 15

 
Maxim Dmitrievsky #:
What do you want, to understand an infinite world or something?

I need to transfer my knowledge to the machine, I think. There is no clear algorithm, there is a set of signs, so I give them to reveal statistical advantage in different situations. I can't trade hands myself - I break the rules - I'm emotional.

No, of course it is good if there will be new patterns, especially working on different instruments.

However, even 4 indicators are enough to memorise a meagre sample - I see a big risk of tinkering.

In your case, how many bars/examples in the history on training? Do you train a set of indicators only once or is there an over-seeding? What is the depth of the tree and how many of them are in the above model? Is the number of quantisation splits set by default?

 
Renat Akhtyamov #:

The branch question is certainly an interesting one....

That's why I was wondering.

Perhaps a pattern can be identified.

I suggest analysing several bars in a row, for example 3-4.

Then move one bar from the beginning of this sample of 3-4 bars and analyse again.

As if overlaying one sample on another.

It is possible to find a pattern

Like this:


In essence, you propose to look for the outcome of the next bar, i.e. how the price will change after a fixed time interval. Then take the results of the model, make some step and train again, adding the results of model classification to the predictors.

 
Aleksey Vyazmikin #:

In essence, you propose to search for the outcome of the next bar from the current one, i.e. how the price will change in a fixed time interval. Then take the results of the model's work, make some step and train again, adding the results of the model's classification to the predictors.

Yeah.

That is, calculate the main pattern that will explain to the trading system the relationship between neighbouring bars and minimize chaos.

 
Renat Akhtyamov #:

Yeah.

I mean, to figure out the underlying pattern

You could just try feeding different samples to continue learning on new data. I think even CatBoost can do that. It also knows how to merge patterns, but I haven't looked into it.

 
Aleksey Vyazmikin #:

You can just try to feed different samples to continue learning on new data. Even CatBoost seems to be able to do this. It also knows how to merge models, but I haven't looked into it.

If you mean different at all, that's not it.

time-shifted on the same data, that's different.

the goal is to determine the relationship between neighbouring bars

 
Aleksey Vyazmikin #:

I need to transfer my knowledge to the machine, I think. There is no clear algorithm, there is a set of signs, so I give them to identify statistical advantage in different situations. I can't trade hands myself - I break the rules - I'm emotional.

No, of course, it is good if there are new patterns, especially those working on different instruments.

However, even 4 indicators are enough to memorise a meagre sample - I see a big risk of tinkering.

In your case, how many bars/examples in the history on training? Do you train a set of indicators only once or is there a seed set? What is the depth of the tree and how many of them are in the above model? Is the number of quantisation splits set by default?

In a normal situation seed has almost no effect, it is the algorithm that matters. If you have to mess with seed, the data is already rubbish

checking on new data solves, if there are only 10 signs, not 1000, you can be sure about it to some extent.

I think the default depth is 6, it doesn't affect much either, except for critical values.

learning depth affects differently, depending on historical variability.

 
Renat Akhtyamov #:

Yeah.

that is to calculate a basic pattern that will explain to the trading system the relationship between neighbouring bars and minimise chaos.

burn yourself out.

 
Maxim Dmitrievsky #:

burn yourself

You're not calmed down, are you?

You're such a bully.

;)))

 
Renat Akhtyamov #:

You're not calming down, are you?

You're such a bully.

;)))

just burn

 
Maxim Dmitrievsky #:

just burn

;)