Train NN via the cloud network

New comment
 

Let's say i am training a neural net .
I have 15000 samples.
These samples are not going to change until the next "harvesting" of data let's say.
Can the cloud network work in a way where 15000 agents can process one sample each in the neural network AND send back adjustment data .

Can it be done ?

(note there are no inputs , the features in the samples are the inputs and they are not ranges but unique fixed values)

Neural Networks: From Theory to Practice
Neural Networks: From Theory to Practice
  • www.mql5.com
Nowadays, every trader must have heard of neural networks and knows how cool it is to use them. The majority believes that those who can deal with neural networks are some kind of superhuman. In this article, I will try to explain to you the neural network architecture, describe its applications and show examples of practical use.
 
Lorentzos Roussos: Let's say i am training a neural net. I have 15000 samples. These samples are not going to change until the next "harvesting" of data let's say. Can the cloud network work in a way where 15000 agents can process one sample each in the neural network AND send back adjustment data. Can it be done? (note there are no inputs , the features in the samples are the inputs and they are not ranges but unique fixed values)

The Strategy Tester does allow you to optimise based on Maths Calculations instead of actual trading optimisations.

I've never used it myself, I don't really know what you are trying to achieve, but maybe that can be used for training your Neural Network?


Math calculations — in this mode the tester does not download history data and information on symbols, as well as does not generate ticks. Only functions OnInit(), OnTester() and OnDeinit() are called. Thus a tester can be used for various mathematical calculations where the selection of parameters is required.
 
Lorentzos Roussos: (note there are no inputs , the features in the samples are the inputs and they are not ranges but unique fixed values)

Maybe you can have a single input, which will serve as the index 1...15000 into the samples.

I don't really know much about Neural Networks, so I am only speculating.

 
Fernando Carreiro #:

Maybe you can have a single input, which will serve as the index 1...15000 into the samples.

I don't really know much about Neural Networks, so I am only speculating.

Thanks

In a simplified approach the samples can be "In the EA" , the network is a matrix of doubles essentially.

So what the user needs in that optimization is the adjustment of the weights resulting from that one sample's lets say "analysis".

So a network of 10 layers with 10 nodes on each layer would have 900 weights . Its these 900 weight adjustments (900 doubles) that each agent would send back ideally.

Then i collect the 900 values from each agent and adjust the network . So there is no need for profit or stats etc or even the genetic algorithm.

I may be butchering terms here like "send back" because i dont know how it works
 
Lorentzos Roussos:

Let's say i am training a neural net .
I have 15000 samples.
These samples are not going to change until the next "harvesting" of data let's say.
Can the cloud network work in a way where 15000 agents can process one sample each in the neural network AND send back adjustment data .

Can it be done ?

(note there are no inputs , the features in the samples are the inputs and they are not ranges but unique fixed values)

Yes it can be done. With Frames.
 
Alain Verleyen #:
Yes it can be done. With Frames.

And i will receive files back ? 

This method needs no genetic algorithm has no reward value needs no stats its just a lot of samples and their "analysis" resulting in the adjustments needed.

To simplify : If i manage to do it in my tester does it mean it can happen on the cloud ?

Thank

New comment