Machine learning in trading: theory, models, practice and algo-trading - page 2289

New comment
 

You can generate long synthetic series, respectively

it is interesting that a small shift of the average in the initial series gives a very strong trend effect, if we generate a lot of data

It turns out that such a model would work only on the real market with an obviously shifted average.

But this can easily be corrected by obtaining a series with the same regularities, but decreasing


 

Trained on generated ones, tested on real ones... so far as an experiment

Sometimes it works. Requires thinking about how to generate, for what cases, etc.


 
Maxim Dmitrievsky:

Trained on generated ones, tested on real ones... so far as an experiment

Sometimes it works. Requires thinking about how to generate, for what cases, etc.


Funny, it should help with re-learning

Is this an autoencoder? How would the latent vector be opened, what row characteristics it has memorized, etc.
 
Rorschach:

Cool, it should help with retraining

Is that an autoencoder? How would the latent vector open up, what row characteristics does it memorize, etc.

Gaussian mixes

I think this simulation is more suitable for martingales and evaluating the stability of axes

 

I will tell you a little bit about how I see regularities in the market

There is a certain pattern (starting point), and then a series of events (rules) happens as a result of which we get a result (Y)

Two-dimensional data

1) extremum (S = suport R = resistance)

2) price of the extremum


Initially data looks like this

price lab
 1.0   R
 0.0   S
 0.4   R
 0.0   S
 0.3   R
-0.3   S
-0.1   R
-0.5   S
-0.1   R
-0.5   S

initial pattern

data for research

I used the SPADE algorithm (helpfully on the wiki) , in order to do that we need to convert the data into a slightly different format, like the event format

[1] "(-0.2)S"  "(2.2)R"   "(1.1)S"   "(3.1)R"   "(2.2)S"  
 [6] "(2.8)R"   "(1.2)S"   "(2.5)R"   "(1.9)S"   "(3)R"    
[11] "(2.4)S"   "(5.1)R"   "(3.4)S"   "(4.5)R"   "(4.1)S"  
[16] "(4.5)R.1" "(4)S"     "(5.3)R"   "(4.8)S"   "(7.3)R"  
[21] "(4.9)S"   "(6.2)R"   "(3.9)S"   "(5.5)R"   "(4.9)S.1"
[26] "(5.7)R"   "(4.8)S.1" "(6.2)R.1" "(4.8)S.2" "(5.5)R.1"
[31] "(4.2)S"   "(5.7)R.1" "(4.9)S.2" "(6.6)R"   "(6)S"    
[36] "(7)R"     "(6.1)S"   "(8.5)R"   "(7.6)S"   "(8.2)R"  
[41] "(7.6)S.1" "(8.3)R"   "(7.8)S"   "(8.4)R"   "(7.6)S.2"

It's basically the same thing, but in a different form.


I run the algorithm, look for the rules...

The algorithm finds very strong rules...

I'll show you one...


The pattern

price lab
0.4   R
0.0   S
1.0   R

Then comes a series of events after which the result...

 "(-0.3)S"   "(-0.6)R"   "(-0.6)R.1" "(-0.6)R.2"


This is a fundamentally new approach to searching for patterns in the market, SPADE has many drawbacks and limitations, and I'm already thinking of another self-written algorithm to search for rules ...

Such non-trivial ideas and tasks...

 
Hasn't anyone written a martingale on neural networks? Googling gives 0 results
 
Maxim Dmitrievsky:
Hasn't anyone written a martingale on neural networks? Googling gives 0 results

That defeats the purpose of the AI

 
Maxim Dmitrievsky:
Hasn't anyone written a martingale on neural networks? Google gives 0 results
Neuronkey's skull is cracking ;)
 
Vitaly Muzichenko:

Then the point of AI is lost.

post

I mean to teach trading by martinOM, not to average trades after training

 
Maxim Dmitrievsky:

post

I mean to teach exactly how to trade by martin, not to average trades after training.

Well, I can't even imagine where to start and how it should look like. In my opinion, this is kind of incompatible things.

1...228222832284228522862287228822892290229122922293229422952296...3492
New comment
Reason: