Machine learning in trading: theory, models, practice and algo-trading - page 2296

 
Maxim Dmitrievsky:

It's fun to play with this, I'll do it soon.

It's funny because it's hard to understand.)

By the way, grids seem a bit more meaningful (compared to martingale). For example you can try to grasp the effect when small movements are more likely to continue and large ones are more likely to reverse.

 
Maxim Dmitrievsky:

For some reason everyone is used to calling martin the averaging of positions, mostly not meaningful or over-optimized

you can leave the lot fixed, but still use a grid. Trade sets (entry and exit points) will change, so will their representation in the feature space for MO. This is exactly the interesting point.

I don't know if the stability effect will appear on new data. I don't have such a mathematical formula. Check it empirically.

Grid and Martin are fundamentally different things in terms of "mathematicsof strategies". ) game theory, if you will.

The grid is only needed to diversify the risk of entering earlier/later than the "ideal" moment, or to make a discount to the "confidence" in entering the market. At the same time, the grid in the plus-pyramid improves no less than the averaging in the minus.

And Martin is exactly a mathematically separated type of averaging with increasing, in which everything is tied to mathematics (there is no need to think and look for inputs, the poor invented filters), and the only question is the thickness of the deposit, how much it can withstand a reverse movement.

About the grid - I like your idea very much. Why should the target function depend on separate trades, if trades themselves can be built with some grid with cunning MM, such as minimum risk for entry and further pyramiding or averaging? It suddenly became obvious that the targets will be different for different entry strategies.

 
Aleksey Mavrin:

The grid and Martin are fundamentally different in terms of the "mathematicsof strategies," if you will. ) game theory, if you will.

The grid is only needed to diversify the risk of entering earlier/later than the "ideal" moment, or to discount the "certainty" of entering the market. At the same time, the grid in the plus-pyramid improves no less than the averaging in the minus.

And Martin is precisely a mathematically separated type of averaging with increasing, in which everything is tied to mathematics (there is no need to think and look for inputs, the poor situation makes up filters), and the only question is the thickness of the deposit, how much it can withstand a reverse movement.

About the grid - I like your idea very much. Why should the target function depend on separate trades, if trades themselves can be built with some grid with cunning MM, such as minimum risk for entry and further pyramiding or averaging? Suddenly it became obvious that the targets will be different for different models with different strategies of dividing the entries.

Well, yes, just a grid. Probably with a slight martin.)

 
Aleksey Nikolayev:

Here, by the way, grids seem a bit more meaningful (compared to martingale). You could, for example, try to capture the effect where small movements are more likely to continue and large movements are more likely to reverse.

Yeah, kind of like that...interesting how MO will generalize to new data. In theory, it should be less sensitive to noise.

 
Maxim Dmitrievsky:

articles are good for the minds of people


No one is interested yet. Hypothesis - the more people know about MO, the less interest in freelancing))


 
Aleksey Mavrin:

No one is interested yet. Hypothesis - the more people are shaky in the MO, the less interest in freelancing ))

Here only 1.5 people can do all of the above )

To be honest, I don't understand how people manage to make profitable bots (e.g. for Market) by simple means a la sketched indicators, optimized and works!

Maybe, the sample is large and someone will be lucky.

 
Awesome, how come I didn't read it before, so many bikes reinvented
Прогнозирование временных рядов при помощи экспоненциального сглаживания
Прогнозирование временных рядов при помощи экспоненциального сглаживания
  • www.mql5.com
Статья знакомит читателя с моделями экспоненциального сглаживания, использующимися при краткосрочном прогнозировании временных рядов. Помимо этого затрагиваются вопросы, связанные с оптимизацией и оценкой результатов прогнозирования, приведены несколько примеров в виде скриптов и индикаторов. Статья будет полезной при первом знакомстве с принципами прогнозирования на базе моделей экспоненциального сглаживания.
 
Rorschach:

It seems to be rap with a lowered bitrate, until I found the link

https://www.mql5.com/ru/forum/143224/page30#comment_3620287

Как отличить график FOREX от ГПСЧ?
Как отличить график FOREX от ГПСЧ?
  • 2013.02.11
  • www.mql5.com
Берется Excel и с помощью функции строится псевдослучайный ряд...
 
Rorschach:

https://www.mql5.com/ru/forum/143224/page30#comment_3620287

It doesn't beat the timeline there - Marsaglia tested his diode-rap noise on his tests in 1995 (the release date of his CD-ROM with that noise), and the NIST test set dates back to 2010. I have faith in rappers, though - they'd probably win the AES contest, too)

Clearly, you can always make a non-random sequence for any set of tests that they don't recognize - it's easiest to take something like the binary notation of pi. Nevertheless, I'd recommend that tsosnickers look into these tests, and not just run around for years with ideas on how to fit a bunch of sinusoids to the market)

NIST Special Publication (SP) 800-22 Rev. 1a, A Statistical Test Suite for Random and Pseudorandom Number Generators for Cryptographic Applications
NIST Special Publication (SP) 800-22 Rev. 1a, A Statistical Test Suite for Random and Pseudorandom Number Generators for Cryptographic Applications
  • csrc.nist.gov
This paper discusses some aspects of selecting and testing random and pseudorandom number generators. The outputs of such generators may be used in many cryptographic applications, such as the generation of key material. Generators suitable for use in cryptographic applications may need to meet stronger requirements than for other applications. In particular, their outputs must be unpredictable in the absence of knowledge of the inputs. Some criteria for characterizing and selecting appropriate generators are discussed in this document. The subject of statistical testing and its relation to cryptanalysis is also discussed, and some recommended statistical tests are provided. These tests may be useful as a first step in determining whether or not a generator is suitable for a particular cryptographic application. However, no set of statistical tests can absolutely certify a generator as appropriate for usage in a particular application, i.e., statistical testing cannot serve as a...
 
Aleksey Nikolayev:

The timing doesn't beat there - Marsaglia tested his diode-rap noise on his tests in 1995 (the date of his CD-ROM with that noise), and the NIST test set dates back to 2010. I have faith in rappers, though - they'd probably win the AES contest, too)

Clearly, you can always make a non-random sequence for any set of tests that they don't recognize - the easiest thing is to take something like the binary notation of pi. Nevertheless, I'd recommend tsosnikov to study these tests, and not just run around for years with ideas how to attach a bundle of sinusoids to the market)

By the way, I remembered reading about a cool experiment - when the convolutional network began to hype the recognition of SEALs and the other 1000 classes,

someone did some research and wrote a simple algorithm that changed almost any picture and tricked this neuronku (AlexNet or something),

And the changes were not visible to the eye, but the grid was broken, something like a shift of half a pixel, etc.