Machine learning in trading: theory, models, practice and algo-trading - page 715
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
I and a couple of other participants have repeated this more than once, but everyone's attention is directed to another grail library (package, parameter configuration), as before to another miracle-industrial in the style of JMA, etc.
Those whose hands don't grow out of their... Those whose hands don't grow straight away have already understood that one can't get prediction accuracy higher than 55% for 1 min, while it's normally 52-53% and the correlation with (Close-Open) the next candle is about 0.05 (R^2 = 0.0025), moreover this forecast is very noisy, while averaging takes all the advantages, but this is reality, the truth which has to be adjusted to. Personally I don't know how to do it yet((( I don't have a good strategy.
You see, what's important here is the approach to the market as a whole. If it's correct, then the TS will not make you wait long. Read my article..... its main point is to show the approach to the market..... If you approach it correctly, you can get quite an interesting advantage. And those who agoldelo throw themselves on kotir as on BP unsteady...... They are usually left with what they had in the beginning. You need to take a broader look at the market. See what options are, how expectations are formed, how they are subsequently traded and how the price reacts after that. This seems to me one of the most correct approaches.... IMHO
Which is exactly what I needed to prove. This is the OOS from the first of February. The number of deals is under 60, and the balance curve did not see any particular drawdowns. Amazing happiness. I need more tests...
And here's one with a 500 pips stop....
And here I removed the "I don't know" signals and left ONLY the signals that the AI is completely confident in. It's a nice picture. And it's all on OOS, in case you haven't figured that out yet....
So who has the grail there????????!!!!! Need more tests.....
Bummer It turned out to be a training plot. I'm redoing it now.... Let's see how it goes...
And here I removed the "I don't know" signals and left ONLY the signals that the AI is completely confident in. That's quite a picture. And it's all on OOS, in case you haven't figured that out yet....
As far as I remember, Reshetov himself said to do this - use only the signals the model is confident in. He even switched to a committee of two models for the sake of that, to get a ternary classifier with them. You should have abandoned the "I don't know" signals from the very beginning.
As far as I remember, Reshetov himself said to do so - to use only signals in which the model is sure. He even switched to a committee of two models to get a ternary classifier with them. You should have given up on "don't know" signals from the beginning.
I will... I'm currently training a model the OOS of which begins in the beginning of February, let's see how it reacts this month...
Not as rosy as the previous post, but 100% OOS
And this one without the stop....
Mihail Marchukajtes:
In this tester and optimization mode, you cannot use stops and takes, or open/close/change trades in the middle of a bar. All trading operations in the Expert Advisor should be performed only at the beginning of the bar.
The terminal should be connected to the trade server on all trading days without interruption, in order to collect the correct bar history. If the terminal is switched off and misses some bars, it will download what the dealing will give it. And dealing in 95% gives nothing instead of real ohlc.
Any violation of these rules will lead to the wrong result.
It is better to transfer the strategy to MT5, it tests quite truthfully there in real ticks mode.
The old tradition is to insert real ticks of Dukascopy broker into MT4 via tickstory, after that the testing will be much more believable.
In this mode of tester and optimization, you cannot use stops and takeaways, or open/close/change trades in the middle of a bar. All trading operations in the Expert Advisor should be performed only at the beginning of the bar.
The terminal must be connected to the trade server on all trading days without interruption, in order to collect the correct bar history. If the terminal is switched off and misses some bars, it will download what the dealing will give it. And dealing in 95% gives nothing instead of real ohlc.
Any violation of these rules will lead to the wrong result.
It is better to transfer the strategy to MT5, it tests quite truthfully there in real ticks mode.
The old traditionally, you can use tickstory to insert real ticks of Dukascopy broker to MT4, after that the testing will be much more believable.
I work only on the first bar. Therefore, the results of the tester are reliable.... 100%
You see, what matters here is the approach to the market as a whole. If it is correct, then the TS will not keep you waiting long. Read my article..... its main point is to show the approach to the market..... If you approach it correctly, you can get quite an interesting advantage. And those who agoldelo throw themselves on kotir as on BP unsteady...... They are usually left with what they had in the beginning. You have to take a broader look at the market. See what options are, how expectations are formed, how they are subsequently traded and how the price reacts after that. This seems to me one of the most correct approaches.... IMHO
I read it, it's as ridiculous as this post of yours.
I read it, it's as ridiculous as this post of yours.
Oh, well. Criticism is a virtue..... Way to go.....
Just tell me, what was so ridiculous about my post? What's wrong with it????
Mikhail, how did the entropy/non-entropy experiments end?