Do you know how to make canals? - page 5

 
Maxim Dmitrievsky:

I posted the results of one such system in the MoD thread. It is dumb but retrains very quickly at every bar, though it is not necessary if the training sample is not 3 bars, because each additional bar is very little new information and has almost no effect on the training result

as the result, an unpleasant thing happens - as soon as enough historical patterns are available for effective prediction - they suddenly change and the system won't work until it has enough examples with new patterns, and so on in a circle.

I.e. NS with retraining works the same way as a lagged muving and does not extract any new "knowledge"

If I want to make a really cool thing using the NS I should use unconventional thinking and geekiness, or teach an already working strategy.

Abandoned NS.

There are characteristic shapes for any chart, for any instrument from which trade decisions are easily found.

 
Uladzimir Izerski:

Abandoned NS.

There are characteristic forms for any chart, for any instrument, from which decisions about trades are easily found.


But I want to make a smart robot :)

 
Maxim Dmitrievsky:

But I want to make an intelligent robot :)


A man can't trade himself, but a robot can))

 
Maxim Dmitrievsky:


As a result, an unpleasant thing happens - as soon as there are enough historical patterns for effective prediction, they suddenly change and the system does not work until there are enough examples with new patterns, and so on in a circle.


Maxim, what kind of historical patterns do you suddenly change?
 
Alexander_K2:
Maxim, what kind of historical patterns do you have that suddenly change?

almost any

it's tantamount to the fact that the system keeps getting into different environments hitherto unknown
 
Maxim Dmitrievsky:

almost any

:))) I don't know what can change there... I have averaged historical variance and coefficient of asymmetry for the distributions almost unchanged.
 
Alexander_K2:
:))) I don't know what can change there... I have averaged historical variance and coefficient of asymmetry for the distributions almost unchanged.

well i write more scalper bots

one year the market goes up - the next year it goes down... the patterns are different

or the volatility also changes from year to year... in crisis years, the market may fall by 100 points every day, while in other years it may fluctuate... but the transactions are in the thousands

 
Maxim Dmitrievsky:

But you want to make a smart robot :)


Everyone dreams, and some are even good at it.

Maxim Dmitriev:

If a person cannot trade himself, a robot will be able to do it))


A robot is not subject to emotions and acts strictly according to the algorithm.

Not everybody can create the right algorithm, not everybody is a composer))

 
Uladzimir Izerski:



The robot is not subject to emotion, it acts according to an algorithm.

Not everyone can create the right algorithm, so not everyone is a composer either))


an algorithm is an algorithmic robot
and we're talking about ns, self-learning systems

 
Uladzimir Izerski:

Everybody dreams, and some people are even good at it.


What's not to get, everything works out... eventually, if you don't give up and brokers don't fuck with you... )

Reason: