Discussion of article "Neural Networks Cheap and Cheerful - Link NeuroPro with MetaTrader 5" - page 8

[mtyvnel]

Thank you very much ! You have helped me a lot ... Now I can not do without this programme ! I just added this code to all my EAs and when I sit down together everything works ! Why do people think it is a tweak ? After all all trading systems are also a tweak in some way ?

Success to you !

[Ivan Butko]

Ivan Butko #:
Just checked: stupid fitting. After a trained period, immediately, right from the next bar, there is a chaotic picture with no signs of life.

Somehow it turned out emotionally. Thanks to the author for the article. At least it is very accessible and quite entertaining to spin the net.

[Ivan Butko]

I'm tired. I've tried all the layers, even all 100 neurons. Even all 512 fields. The computer almost exploded, and the result is 50/50, a balance flat. Hitting a properly trained network is like flipping a coin. The price breaks right after the backtest, unfortunately. Pricing is an untrainable chaos of values. Apparently, it is not a neural network that is needed here, but something else.
But I was curious to twist and turn the program.

[Roman Shiredchenko]

Colleagues - I think I read it for 2 times. I will edit and look under my desires. .... first in the basic version.... where do we tell the grid, in what criteria - that it should provide the value of the closing price of the current zero bar - hour in the output layer?

[Эдуард]

Kind of typos in the script code

1)

if(i==0) row+=TimeToString(rate.time,TIME_DATE || TIME_MINUTES)+",";

I think there should be 1 operand |.

[Эдуард]

The network has shown itself well on training, but price charts have no regularities (completely random) and it makes no sense to use indicators or networks for forecasting.

[Ivan Butko]

Эдуард #:
The network has shown itself well on training, but price charts have no regularities (completely random) and it makes no sense to use indicators or networks for forecasting.

What is surprising, having trained the network to grail even for 20 years (every candle the neural network knows where to open), on the forward literally from the very first candle - the network breaks down, 50/50.

It would seem that 140,000 candles are behind us, there will probably be patterns on the forward.... but no. What a miracle of non-stationarity.

Moreover, I exported all currencies for the last year to one dataset (40 years of "fresh" history in total, i.e. each currency pair has only the last year), I thought, well, now it will play for sure, a universal neuron that can trade on all currency pairs.

The result: the grail on all currencies for a year of backtesting, and forward.... from the first candle 50/50.

Well, at least for a month, at least for a week, at least for a day, but trade steadily!..... Yeah, no way.

[Renat Akhtyamov]

It's simple

You open a trade, the price goes the other way, against the crowd or against the maximum risk.

What neuro can counteract that?

It's useless.

[Victor Castillejo]

Good evening andrew, I can't get the NeuroPro software, can you tell me where I can get it, please....

[Dmitrii Shershov]

Советник на этом примере скомпилировался, но при тестировании выходит ошибка "array out of range in" и советник закрывается.

On debugging it shows the line - BAR[bar]=rate.close-zlevel;

Who can tell me what is the reason?

Who can tell me what is the reason?