Machine learning in trading: theory, models, practice and algo-trading - page 3108

 
Maxim Dmitrievsky #:
A properly asked question usually already contains an answer. Apparently such a question has not yet been asked.

Perhaps you should ask yourself what is similarity as a whole and what clustering has to do with it.

If you need to estimate the density of distribution (I am trying to guess from the inexplicit question), then it is kernel density estimation.

If you sincerely wish to help, I will burden you with the details of the problem.

So, we have a conditional formula:

P=A/(A+B)*100-x.

Where, A and B are integers, let's say from 1 to 1000.

x is some coefficient, even let it have a fixed value for simplicity, and it can be removed in the mind.

There is a sample, let's say 500 rows.

For each row we calculate the value of P. We split the results into ranges by one or another method, and as a result we have an empirical distribution.

However, the formula itself gives comparable values at different values of A and B, which is logical, because in essence we count the percentage of the population, so it is important for me to make a correction for the number A, and for this purpose another space is made, where the value of A in the formula is specified, and the empirical diagram is also built.

This produces a 3d diagram, which can be written as a matrix.

I have about 10 thousand such matrices, and I want to group them by similarity.

So I need a clustering method that will gather them into groups. Of course, I would like them to be similar not just by individual points, but also by distributions.

For example, we can compare distributions of each layer (partitioning by the number A) and then calculate the percentage of similar distributions. But, maybe there is a good solution ready.

Have I described the essence of the problem clearly?

 
That's how to find the optimal number of clusters. You can try a Gaussian mixture instead of kmins.
 
Aleksey Vyazmikin #:

However, who am I talking to - a lover of off-the-shelf solutions.....

Aleksey Vyazmikin #:

Let's say you can compare distributions of each layer (partitioning by number A), and then calculate the percentage of similar distributions. But maybe there is a good solution ready.

How can you do that? )))

 
mytarmailS #:

How can that be? )))

like he caught you and gloats))))))

Those who can create custom solutions can also use ready-made ones. But the other way round - no.

So you didn't catch me.

 
Andrey Dik #:

kind of caught him and he's gloating.)

Those who can create custom solutions can also use ready-made ones, but the other way round - no.

So you didn't catch me.

There is such a phenomenon as hyperactivity, when a person seems to be everywhere and at the same time nowhere. Similarly, uncontrolled use of packages leads rather to fatigue and disorientation, and then to anger, than to gaining any minimal knowledge :)
 
Maxim Dmitrievsky #:
There is such a phenomenon as hyperactivity, when a person seems to be everywhere and at the same time nowhere. In the same way, uncontrolled use of packets leads to fatigue and disorientation, and then to anger, rather than to the acquisition of any minimal knowledge :))
you are an optimist Max, I think it's much worse))).
 

I tested the bot that I threw here on the 15th of May, a month has passed. With different sl and tp different results, but on average all on the growth.


 
Maxim Dmitrievsky #:

I tested the bot that I threw here on the 15th of May, a month has passed. With different sl and tp different results, but on average all on the growth.


Take 5 pips stop 50 ?
 
mytarmailS #:
Take 5 pips stop 50 ?

it's a long term pattern, the signal stays in one direction for a long time.

I've shown that for 20 years, only 1000+ trades. Short takeouts are one of the ways out , otherwise there will be few trades.

 
Maxim Dmitrievsky #:

is a long-term pattern, the signal persists for a long time in one direction

showed that in 20 years, only 1000+ trades. Short takeouts are one of the ways out , otherwise there will be few trades.

It's some kind of tactic I'm not familiar with.

Like a buy signal is long term, but the buy positions themselves open and close often?
If so, why not just hold the buy?