Neural networks. Questions from the experts. - page 12

New comment
 
I did so ; I put the test report into Excel, deleted missing values from the balance and profit columns to get homogeneous columns, then put it into the forecasting software and the output was a forecast of future trades ...If you can show that it is not so I will be glad to see.
 
Urain писал(а) >>

(didn't seem to be already :o)


It was. VictorArt. He trades by trade history. The motto is stability. Stability of what indicator he does not specify.
Here is his PAMM.

 
It would be good to know the principle or method he uses. If you can, please tell me where to read it, because it's not googling it.
 
AAAksakal >>:
Хорошо бы знать принцип или метод который он использует . Если есть возможность то то подскажите, где можно почитать.А то, что то не гуглит .

https://www.mql5.com/ru/forum/124583/page2

 

Question: How do I feed an "empty" value into a network (PNN in particular)?

Suppose a vector {1, 0, 0, 1, 1} is given as input, where

1st input 1 - event A1 happened, 0 - event B1 happened.

2nd input 1 - event A2 occurred, 0 - event B2 occurred. etc.

What if none of the events have occurred in any element of the pattern (vector)? How to correctly represent a vector {1, 0, 0, NULL, NULL}?

 
lasso:
....

What if none of the events have occurred in any element of the pattern (vector)? How to correctly represent a vector {1, 0, 0, NULL, NULL}?

%)
 
lasso:

Question: How to input an "empty" value into a network (particularly PNN)?

Suppose a vector {1, 0, 0, 1, 1} is fed to the input, where

1st input 1 - event A1 happened, 0 - event B1 happened.

2nd input 1 - event A2 occurred, 0 - event B2 occurred. etc.

What if none of the events have occurred in any element of the pattern (vector)? How to correctly represent vector {1, 0, 0, NULL, NULL}?

Filter such shit BEFORE feeding it into neural network.

Otherwise you will get crap instead of training.

 

Ah... I looked more at VictorART's reasoning on the forum.

Basically, after the words "If the market changes so dramatically that the old trading robots lose effectiveness, you can always quickly and automatically create new adaptive trading robots more adapted to the new trading conditions " you don't have to read any further )

 
Diamant:

Filter that shit out BEFORE you feed it into the neural network

Good thing they didn't write "... BEFORE submitting it to the forum" )))

I'll try to describe it in another way.

..........

There is a signal S1, say, for Buy, after that we start monitoring if the price reaches certain levels,

For example, we define levels +/- 20, +/- 30, +/- 50, +/- 90, +/- 120 and if price reached +20-pp, then +30-pp, then rolled back 100pp and reached -50-pp, then -90-pp, then went up again to +120-pp, thus we have a record of how this signal S1= {1, 1, 0, 0, 1} and so on.

But if some signals are very close in time, the price may not reach some levels, so for some signal we have Sn= {1, 1, 0, }, i.e. +/- 90 and +/- 120 levels did not work.

But nevertheless it is not a shit, it is also a training example for NS (showing that the signal was weak, short-lived, or whatever it may say...) and it would be nice to present it for training too, not to filter it out.

The first thing that comes to mind is the 0.5 value for the level that did not trigger. But for some reason I don't think that's right. What is the correct value?

 
lasso:

For example, let us define levels +/- 20, +/- 30, +/- 50, +/- 90, +/- 120 and if price reached +20-pp, then +30-pp, then rolled back 100pp and reached -50-pp, then -90-pp, then went up again to +120-pp, so the record of such signal returning is S1= {1, 1, 0, 0, 1} and so on.

But if some signals are very close in time, the price may not reach some levels, so for some signal we have Sn= {1, 1, 0, }, i.e. +/- 90 and +/- 120 levels were not triggered.


....... But for some reason I don't think that's right. What's the right thing to do?

Enter the third type of signal. Total signals:

0 or 1 or 2

1...5678910111213141516171819...25
New comment