Machine learning in trading: theory, models, practice and algo-trading - page 2582

New comment
 
Maxim Dmitrievsky #:
This is when you know what to create and why. They are not ready-made, the transactions are sampled randomly, like in the articles. There are no a priori assumptions or heuristics at any stage of data preparation, there are some value ranges like maximum and minimum deal holding times.
Okay, I don't want to argue...
I just recommend training something with a fitness function and you'll see how much more you can do with this approach
 
mytarmailS #:
Okay, I don't want to argue...
I just recommend training something with a fitness function and you'll see how much more you can do with this approach
I did so back in the early days of MO, through MT5 optimizer and selection of weights, but there are few tools to fight with fitting in this variant. You set optimization criterion, feature space and optimize weights of neurons.
It's similar with fuzzy logic, there's an article. Then development through reinforcement learning, also through optimization criterion. That doesn't work head-on, overtraining.

Complicating the approach is not a fad, just the result of rethinking the ones that don't work. Plus competition and leveling out inefficiencies that used to be there. You can't get anywhere by simple means anymore.
 
Maxim Dmitrievsky #:
I've done so in the early days of MI, through MT5 optimizer and selection of weights, but there are few tools to fight with fitting in this variant. You set optimization criterion, feature space and optimize weights of neurons.
With fuzzy logic is similar, there is an article. Then development through reinforcement learning, also through optimization criterion. It does not work head-on, overtraining.

Complicating the approach is not a fad, just the result of rethinking those that don't work. Plus competition and leveling out inefficiencies that used to be there. You can't get anywhere by simple means anymore.

Okay, like I said, I'm not going to argue, even though I don't agree.

You need to think about TC production, I've never done that, just looking for an algorithm and research...

I need to learn another programming language or learn to work with API, what is better?

 
mytarmailS #:

Okay, like I said, I will not argue, although I do not agree...

We need to think about TC production, I've never done that, only looking for an algorithm and research...

Need to learn another programming language, or learn how to work with API which is better?

It depends on the level of production. If Forex - MT5, I think there is a connector for R, about python I am silent. There is a way to transfer the catbust to MT5. From practice - it is better to convert models to the platform language, without api.
 
Maxim Dmitrievsky #:
It depends on what level of production. If Forex - MT5, I think there is a connector for R, I will not say anything about python. There is a way to transfer the catbust to MT5.

And what is considered a small non-floating spread in forex?

 
mytarmailS #:

What is a non-floating spread on forex that is considered small?

It depends on the instrument, on Eurobucks I think the average was 2p.
 
Maxim Dmitrievsky #:
It depends on the instrument, on Eurobucks I think the average was 2p

It is a lot of pancakes

Ahahahaha, yesterday I read the quotes from MT5 and roboforex, and of course there is a floating spread in the history, I looked and was shocked, every day it jumps from 1 point to 20 points )))) I dropped out of my mind
 
mytarmailS #:

Man, that's a lot.

Ahahahaha, yesterday I read the quotes from MT5 with Roboforex, and of course there is a floating spread in the history, I looked and was shocked, every day it jumps from 1p to 20p )))) I dropped out of my mind

And at night it could be as high as 100

 
Maxim Dmitrievsky #:
Norm, start trading
Is that your wellslab? How do you transfer models there?
By the way, yes, I support it...
 
Mihail Marchukajtes #:
By the way, yes, I support you...

I'm using Wells-lab. I don't have a normal connector for MOEX, but here there is enough room for trading in MT5.) I worked with models in Vels: I have human models on well-known frameworks, API in Python, to which I send a request with the vector of fields and model name. And API loads all models into RAM during initialization. Gets the request, finds the model, feeds the data, gets the prefix, returns the prefix. Everyone is happy. Perhaps I will use the same scheme in MT5, moreover the half of this design may be the same, only MT5 part should be implemented.

1...257525762577257825792580258125822583258425852586258725882589...3590
New comment