Searching for an arbitrary pattern using a neural network - page 6

New comment
 
Peter, after familiarising yourself with the nets, have a look at the convolution nets.
 
Реter Konow:

It is one thing to recognise by price data, it is another to recognise by colour. Still, completely different approaches and mechanisms.

Do you think it makes sense for a PC to have a colour chart scheme? )))

OK, I give up, otherwise you'll continue to make me laugh ))))

For NS and indeed for any algorithm of interaction with PC all data will be presented in the form of arrays (memory or arrays are not important here)

and it won't make any difference what you teach NS to OHLC arrays, what you teach to bitmask arrays of screenshot,

.....though in machine learning there is a certain "trick" that data and configuration and type of NS may matter - but here more randomness rules ;)

 
Igor Makanu:

Do you think it makes sense for a PC to have a colour scheme of graphics? )))

OK, I give up, otherwise you'll continue to make me laugh ))))

For NS, and indeed for any algorithm interacting with the PC, all data will be presented as arrays (memory or arrays are not important here)

and it won't make any difference what you teach NS to OHLC arrays or screenshot bitmask arrays,

.....though in machine learning there is a certain "trick" that data and configuration and type of NS may matter - but here more randomness rules ;)

No doubt you understand more than I do about MoD, but there is a logical inconsistency here. OCHL data and pattern screenshot data are fundamentally different data at computer level. In the case of price it is double, in the case of colour it is uint. In case of OCHL we need to analyze a correlation of values of price parameters of bars, and in case of a picture - the correspondence to the sought image. Training with OCHL data is a search for numerical, not graphical patterns (which are, understandably, also numbers for a network). Learning graphical patterns, on the other hand, uses an entirely different material and method. Perhaps findinga graphical pattern through a numerical pattern is wrong. I think these are different approaches to learning and recognition.
 
Aliaksandr Hryshyn:
Peter, after familiarising yourself with the nets, have a look at the convolution nets.
I will.
 

Mother of God!


 
And the amount of OCHL data per pattern is ~ 10 or 100 numbers, and a graphic image ~ 300*300 pixel colour values.
 
Dmitry Fedoseev:

Mother of God!


Don't beat yourself up so many times, everyone makes mistakes.)
 
Реter Konow:
Don't beat yourself up so many times, everyone makes mistakes.)

But not everyone sticks their necks out like that.

 
Igor Makanu:

Alas, you are not fixable!

The computer doesn't care what it processes - in the end it doesn't even know what it was given, be it a picture or nuke data or OHLS... numbers are numbers as they are!

I don't know how else to explain that the PC is not smart - it's dumb hardware, what you give it into the algorithm, it will process in the algorithm!

so did you explain it?

))))

Do you think the NS is a "magic wand" to which whatever you give it, you always get what you need? It doesn't matter what the data is, it doesn't matter how big it is. It's all numbers...

Then I do not understand, where is the algorithm that finds patterns? Where is this "almighty" NS? They've been studying MO for so long and still no "pattern recognizer" in MT's arsenal.

 
Dmitry Fedoseev:

But not everyone sticks their necks out like that.

Can you as an expert make an NS that recognises at least 5 patterns on any chart and timeframe?
12345678910
New comment
Reason: