Libraries: RL algorithms - page 2

New comment
 
Igor Makanu:

If primitive methods can find patterns, then machine learning is all the more obliged to find it all.

Maxim Dmitrievsky:

That it approximates some patterns

I think that between these 2=two phrases lies the truth, which will allow to create a workable TS capable of adapting to the market, something similar to the BestInterval library, i.e. RandomForest itself as a source of signals for inputs and outputs, most likely, will be just a fitting on history, and if RandomForest can filter (or adapt) relatively workable TS - this is an interesting area of research.

ZY: time is a problem, I don't have it.

PSPS: somehow I managed to crash my previous post ))))))).


 
Added 2 libraries (variations on a theme). Recursive - simple search of predictors by the smallest Log loss. Trigonometric - using cosine transformations.
 
FxTrader562:

Thanks for uploading :)))

I will suggest you to add a separate function for feature transformation so that you can add as many polynomials as you want

Again and again you don't have to write code inside the RDF :))))))

Something like this I recommended long ago:

Hi can you pls leave full code here :) I'll attach it later

 
FxTrader562:

Hi Maxim,

Do you have any solution to handle large data training?

I mean that if I'm getting more than 10 MB, then the EA doesn't run ...

So my question is "Is there a way to handle such a problem"?


only install fewer trees

RDF have a large files with structure always

stop delete all ))

 
Maxim Dmitrievsky :

only install fewer trees

RDF have a large files with structure always

stop delete all)))

Ok, sorry ... I forgot it is your thread :)))

I will not delete your comments in your thread :)))))

Also, it seems like it's a key to RDF success to train with large data.

Now, I will test it after training for 1 to 5 years of data.

 
Zero divide error in "RL recursive.mqh".....
 
FxTrader562:
Zero divide error in "RL recursive.mqh".....

In this library 1 predictor/another predictor. So maybe you put predictors with zeros

 

Good afternoon.

It may be worthwhile to create an enumeration prop that sets the training method for an agent, respectively a function that allows you to arbitrarily set the training method for a particular agent. (Methods of training in the form of functions in one module). It will be possible to create collections mixed by learning method. How do you see it?

 
Maxim Dmitrievsky :

In this library 1 predictor / another predictor. So maybe you put predictors with zeros

I have not added anything to the EA or library. I just tested the default library and EA which you uploaded and got this error.

Anyway, I will figure it out when I will use it.Now I am not using it:))))))

 
mov:

Good afternoon.

It may be worthwhile to create an enumeration prop that sets the training method for an agent, respectively a function that allows you to arbitrarily set the training method for a particular agent. (Methods of training in the form of functions in one module). It will be possible to create collections mixed by learning method. What is your view on this?

Kind, it is possible, but it's a maze for me.... :) There will still be experiments with different variants, now another one will be added (linear).

12345
New comment