Machine learning in trading: theory, models, practice and algo-trading - page 2732

 
Maxim Kuznetsov #:


SUMMARY: we should pay attention to the "detrending" operation. On the one hand, there is no way to do without it, on the other hand, too much necessary information dies there.

To put it simply, a trend is an illusion of a beginner who got rich on history: he entered here and exited here.

This illusion can probably be supported by the MOE, if it is possible to find a predictor for the "trend" teacher, and the model will take into account the neighbouring values of the signs. Seems like a fairy tale to me.


More realistically, there are statistics that say that financial markets are NOT stationary. There is no choice - we will model non-stationary time series. A trend is the first and obvious sign of non-stationarity. There is no choice, we have to detrend the time series. Since the sign of the next bar is predicted, not its value, the loss of information is not fatal.

 
СанСаныч Фоменко #:

To put it simply, a trend is an illusion of a beginner who got rich on history: here I entered and here I exited.

This illusion can probably be supported by the MOE, if it is possible to find a predictor for the "trend" teacher, and the model will take into account neighbouring values of signs. This seems like a fairy tale to me.


More realistically, there are statistics that say that financial markets are NOT stationary. There is no choice - we will model non-stationary time series. A trend is the first and obvious sign of non-stationarity. There is no choice, we have to detrend the time series. Since the sign of the next bar is predicted, not its value, the loss of information is not fatal.

(I remember) I hope they won't shoot you for the link in CodeBase : https://www.mql5.com/ru/code/36558.

may come in handy in predicting signs - predict to your heart's content :-) the indicator just shows (and summarises) the "black/white" signs.

BlackAndWhite
BlackAndWhite
  • www.mql5.com
Наглядно показывает соотношение чёрных и белых свечей
 
The trend should be understood as a shift of the average increments, if only they are used

No matter how stupid a TS is, it is good if it falls into the trend by the directions of trades
 
СанСаныч Фоменко #:

Read the garch and don't make things up

Read more meaningful texts than some brief dictionaries. Start with Robert Engle's original article from 1982.

In your dictionary, of course, there is also about white noise, which is also Gaussian - it is just called differently there (innovations).

 
Aleksey Nikolayev #:

Read more meaningful texts than some brief dictionaries. Start with Robert Engle's original 1982 article.

In your dictionary, of course, there is also about white noise, which is also Gaussian - it is just called differently there (innovations).

It turns out that your knowledge is not at zero.

So let's take a package, for example, rugarch, and discuss modelling in its terms, which cover all nuances of non-stationary series.

 
СанСаныч Фоменко #:

It turns out that your knowledge is not at zero.

So let's take a package, for example, rugarch and discuss modelling in its terms, which cover all the nuances of nonstationary series.

The package is quite good, no argument. But at the moment I am interested in another type of nonstationarity, like the one that arises in Shiryaev's decomposition problems. One often speaks of piecewise stationarity.

 
Aleksey Nikolayev #:

The package is quite good, no doubt about it. But at the moment I am interested in another type of nonstationarity, like that which arises in Shiryaev's decomposition problems. One often speaks of piecewise stationarity.

Any waste of available designs is called science, which is a very fidgety broad.

The length of a piecewise time series is most likely also a non-stationary series, you can look at the ZZ. We get the same problems, only viewed from the side.

 

who's been preaching something like this?

https://cs.stanford.edu/people/karpathy/convnetjs/demo/rldemo.html

 
mytarmailS #:

who preached something like this?

https://cs.stanford.edu/people/karpathy/convnetjs/demo/rldemo.html

does it work ?

demo about a cockroach ( a circle with legs and eyes), rules " we ' ll start out with something more simple: a 2D agent that has 9 eyes pointing in different angles ahead and every eye senses 3 values along its direction (up to a certain maximum visibility distance): distance to a wall, distance to a green thing, or distance to a red thing. The agent navigates by using one of 5 actions that turn it different angles. The red things are apples and the agent gets reward for eating them. The green things are poison and the agent gets negative reward for eating them. The training takes a few tens of minutes with current parameter settings."

you can click start learning... then stop learning...

the cockroach is supposed to run and prefer red dots, avoiding green dots...

In reality: after stop learning, it more or less follows the last movement pattern and does not distinguish between red and green. Or I've got an unusually stupid cockroach :-)

 
Maxim Kuznetsov #:

Does it work?

demo about a cockroach ( a circle with legs and eyes), rules " we ' ll start out with something more simple: a 2D agent that has 9 eyes pointing in different angles ahead and each eye senses 3 values along its direction (up to a certain maximum visibility distance): distance to a wall, distance to a green thing, or distance to a red thing. The agent navigates by using one of 5 actions that turn it different angles. The red things are apples and the agent gets reward for eating them. The green things are poison and the agent gets negative reward for eating them. The training takes a few tens of minutes with current parameter settings."

you can click start learning...then stop learning.....

the cockroach is supposed to run and prefer the red dots to avoid the green dots...

In reality : after stop learning it more or less follows the last movement pattern and does not distinguish between red and green. Or I've got an unusually stupid cockroach :-)

because you should not press stop learning before the learning is complete .