neural network and inputs - page 35
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
Tested it in practice. But can be repeated for clarity.
Suggest input data (excluding OHLC) and teacher. The task to be solved is classification. I.e. teacher should give +1/-1 signals (buy/sell, up/dn, etc.).
Train 3-4 methods, get numerical metrics of these methods that will show accuracy of each one.
Good luck
Tested it in practice. But can be repeated for clarity.
Suggest input data (excluding OHLC) and teacher. The task to be solved is classification. I.e. teacher should give +1/-1 signals (buy/sell, up/dn, etc.).
Train 3-4 methods, get numerical metrics of these methods that will show accuracy of each one.
Good luck
No need to repeat. Let's get the results of validation checks in practice with all the parameters.
Good luck
I'll get it ready this weekend.
Good luck
I'll get it ready this weekend.
Good luck
Looking forward to it.
Yours sincerely
Guys, I propose - the GURU on the fifth, if not for a conversation, then on the opinion on his part to invite, so that he voiced (wrote), where we move in this direction, and, in general, his opinion!!!
Reshetov please do not worry, with his "bread and sausage", in his words, on his R-Net - "fed up" ... Fortunately not on the real one... Just don't fuck around...
Once in the subject - let him speak (recommend) on his part the direction of travel and development ... All the more so for three hundred bucks.
All, IMHO, just a suggestion... Organize (or post in the existing) on the fifth theme.
From the heart - without uzhey@ons.
Tested it in practice. But can be repeated for clarity.
Suggest input data (excluding OHLC) and teacher. The task to be solved is classification. I.e. teacher should give +1/-1 signals (buy/sell, up/dn, etc.).
Train 3-4 methods, get numerical metrics of these methods that will show accuracy of each one.
Good luck
This is the teacher you should try. (https://www.mql5.com/ru/code/903). It doesn't get any better than this.
The inputs are up to you, even OHLC.
Some thoughts are all imho.
Next we will talk about fixed lot trading, money management is a completely different topic.
It would be logical to evaluate an EA's performance based on equity. Perfect entries are of course very good, but let us be realistic nevertheless. I propose the following evaluation method. The equity slope depends on correctness and frequency of trades. But this is not the only thing as every normal person is interested in how close the real equity is to the ideal one. The results are estimated on the basis of the last N trades, let N=10. Then ideal trade would be 1 1 1 1 1 1 1 1 1. Equity of such trade 1 2 3 4 5 6 7 8 9 10. Suppose the real trade is 1 1 -1 1 1 1 1 -1 1 1. Then its equity will be
1 2 1 2 3 2 3 4 3 4. Now let us calculate the drawdown and for the real equity let us read the linear regression (read the slope angle of the linear regression) and the standard deviation from the linear regression line (and not from the average value). We obtain the second characteristic of equity that characterizes the drawdown. Now if we divide the slope angle by the deviation, we will obtain one value that fully characterizes the trade. Now it can be used as a fitness function for adjustment. Of course, a situation may occur when ideal equity is obtained on the last trades, in which case the dispersion is 0. How to avoid such a situation and renormalize the fitness function to the range +1 -1 is our homework.
In principle, there is another option, to calculate the correlation coefficient between the real and ideal equity. There are, of course, plenty of other ways to do it.
As for nerve-networks, I think they, like homo sapiens, are divided into dowsers, dowsers capable of learning and nerds. Nerds just memorize material without any attempt to comprehend it on the back test show perfect results. Downs here are all clear, if you believe your eyes and ears mostly sit in management and make decisions.
Our goal is the downs capable of learning. I think one of the signs we have the best result in the range of 0.8-0.9 for the fitness function on the back test, the network does not retrain. There is some possibility that this is a found pattern.
I will not repeat my vision of the internal organization of the neural net above. There is a reasonable question what is more profitable to teach down or dumb nerd. In the first case we just need to add neurons and optimize them, in the second case we need to sink neurons like kittens. I think the first option is preferable.
I have written earlier what to feed to the input. I have one more thought I will answer in private if you have a report. In a nutshell, we have fish and we don't have fish.
Some thoughts are all imho.
it is logical to judge the results of any expert's work on the basis of the real and only the real. And this is not even IMHO. If you have 20 Expert Advisors with a positive balance curve in the tester, then evaluating the results of their trading using any methods is intellectual rubbish
Put them all on a real account with minimal deposit and the real will show everything. If 10 out of 20 remain with a positive balance for, for example, three months, then we can do different bending.
As it is not very short, I am enclosing it in an appendix.
Good luck
As it is not very short, I am enclosing it in an appendix.
Good luck