Using neural networks in trading - page 13

[solar]

[[Deleted]]

grell:


the reason is different. network redundancy.

And let's work out what network redundancy is?)

[TheXpert]

Figar0:
And let's work out what network redundancy is?)

Fit in essence.

[Дмитрий]

TheXpert:
A fitting essentially.

Fitting is when the sample is too small. Network redundancy has little effect on fitting.

[Дмитрий]

Figar0:
Let's work out what redundancy means, shall we?)

There's no need to, it's already clear that the number of layers and weights is very large.

But I will add to the question of fitting... A redundant network is like a system of 4 equations with two unknowns. Either the network will trivially learn all the data, or the solution will be correct but unstable.

[[Deleted]]

TheXpert:
A fitting in essence.

This is understandable. How do you determine the necessary sufficiency of a network? If there is redundancy, is there sufficiency?

grell:

Either the network will learn all the data.

How much data can the network learn?

[Дмитрий]

Are we talking about all networks or MLPs?

[[Deleted]]

grell:
Are we talking about all networks or MLPs?

What's the fundamental difference? Let it be MLP. Here is your MLP in your own configuration, how much can it learn, fit?

[TheXpert]

Figar0:

This is understandable. How do you determine the necessary sufficiency of a network? If there is redundancy, is there sufficiency?

Ah, that's easy. As soon as it starts to learn, it is sufficient.

[Дмитрий]

The maximum I have achieved is 3 months. At (k/(l+1))*(m/(n+point)=8, where k-number of profitable trades, l-number of losing trades, m-total balance of profitable, n-total balance of losing trades.