What to feed to the input of the neural network? Your ideas... - page 34

 

Record

I have never seen such a beautiful set, and the top of the list.


Optimisation - year: from 2021 to 2022 on EURUSD



Forward on the same - year: from 2022 to 2023.


A cheater year for the forward, it often repeats the success of the 2021 optimisation, but I have not come across such a smooth and more dense year in terms of trades.

Still 3 inputs, 3 neurons. MT5 optimiser, I have moved away from third-party programs for now.

 
Ivan Butko #:

Record

I've never seen such a beautiful set before, and the top of the list.

Optimisation - year: from 2021 to 2022 on EURUSD

Forward on the same - year: from 2022 to 2023

A cheater year for the forward, it often repeats the success of the 2021 optimisation, but I have not come across such a smooth and more dense year in terms of deals.

Still 3 inputs, 3 neurons. MT5 optimiser, I have moved away from third-party programs for now.

ohtyzhyozhyoshkinkot

 
Ivan Butko #:

Forward on the same - year: 2022 to 2023

What will the forward show from 2023 to 2024 ?

 
Alexander Sevastyanov #:

And what will the forward show from 2023 to 2024 ?

Decline. At the beginning, he held on for a while and then went flat.

 
Uladzimir Izerski #:

Without the influence of news (or FA as anyone prefers), neural networks would work just fine. Very fine, in fact.)

Volodya, the trouble is that news breaks the work of many TS.

 
Ivan Butko #:

Decline. At first, it held on for a while and then went flat

That's sad. Most likely it will be even worse in 2024 - 2025.

 
Alexander Sevastyanov #:

Sad. It's likely to get even worse in the 2024 - 2025 stretch.

Sure. For sure.

2022 is a cheater. You can sell neuronics on it to show that they work

 
Ivan Butko #:

Decline. At first, it held on for a while and then went flat

And before running forward 2024 - was the network retrained on 2023?

 

No, did the usual optimisation in MT5 for 2021.

Then ran forward and the whole of 2022 is higher in the picture, with a behind the scenes decline for 2023.

Systems on the simplest ns can work, but finding the right set is unrealistic. Sometimes there are "upstarts" like the one above. But, on other years, on other currency pairs, optimisation does not improve anything.

 

It is unrealistic to use a year or more of NS without retraining.

If the network can work for a week or two without retraining (a month at most) - this is already an excellent result.

Although some pipsarians retrain the network every 24 hours