Discussion of article "Neural Networks Cheap and Cheerful - Link NeuroPro with MetaTrader 5" - page 4
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
The main problem is the formation of an initial set of variables, and for a specific target variable.
From my own experience.
For the binary target variable "long-short" I managed to pick up input variables, though poorly, but not very well.
For the binary target variable "growth over 10 pips - no growth of 10 pips" nothing can be picked up at all.
I don't do bullshit.
Proof.
The posted results always refer to "out of sample training" data. This is done as follows in Rattle:
1. the original set is divided into three parts: 70-15-15%
2. training is conducted on the 70% part, which is called training.
The main problem is the formation of an initial set of variables, and for a specific target variable.
From my own experience.
For the binary target variable "long-short" I managed to pick up input variables, though poorly, but not very well.
For the binary target variable "gain over 10 pips - no gain of 10 pips" nothing can be picked up at all.
I can give you a tip on how to forecast volatility more easily, not only the direction of trades. However, it works only on timeframes smaller than daily.
Enter three additional binary signs for each session in the sample, i.e. three more columns in the table: Asian, European and American. And mark them with some values: for example: 1 - session has come, 0 - another session. In this case, binary classification forecasts for sideways and trends, will have acceptable sensitivity and specificity.
Moreover, unlike other regressors, the signs of sessions can be placed in the future. That is, if we forecast the volatility of a future bar, we mark it as belonging to the corresponding session. There will be no problems with "peeking" because, unlike volatility, we know the belonging of bars to sessions in advance, i.e. entropy in this case is zero.
I can give you a tip on how to predict volatility more easily, not just the direction of trades. However, it works only on timeframes smaller than daily.
Enter three additional binary signs for each session in the sample, i.e. three more columns in the table: Asian, European and American. And mark them with some values: for example: 1 - session has come, 0 - another session. In this case, binary classification forecasts for sideways and trends, will have acceptable sensitivity and specificity.
Moreover, unlike other regressors, the signs of sessions can be placed in the future. That is, if we forecast the volatility of a future bar, we mark it as belonging to the corresponding session. There will be no problems with "peeking", because unlike volatility, we know the belonging of bars to sessions in advance, i.e. entropy in this case is zero.
Yeah... The range of opinions is depressing.
Many people do not know how easy it is to add R to MQL, some people consider packages developed by scientists from universities of the world to be rubbish, and others compare an ensemble of neural networks and randomForest.
SanSanych, I don't understand what the discussion is about?
Forgive me, where to download the specified neuropro 0.25?
Even the author of the programme doesn't know that:
"The old version of the NeuroPro neuroimitator software, which was freely distributed on the Internet, is now 14 years old. I stopped its support and distribution and I don't know where the distributions could be saved in the Internet (and even if I did - why would I need a lot of questions about the problems of possible incompatibility of ancient software with fresh versions of Windows?)".
Source: http://neuropro.ru/faq.shtml
Forgive me, can you tell me where to download the specified neuropro 0.25?
Yeah... The range of opinions is depressing.
Many people do not know how easy it is to add R to MQL, some people consider packages developed by scientists from universities of the world to be rubbish, and others compare an ensemble of neural networks and randomForest.
SanSanych, I don't understand what the discussion is about?
Forgive me, can you tell me where to download the specified neuropro 0.25?