Machine learning in trading: theory, models, practice and algo-trading - page 2068
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 think I understood you and wrote in the spirit that an event repeating every day at 9 am would also be an event repeating every Wednesday at 9 am. It would be quite difficult to isolate those events that have a REALLY WEEKLY (but not daily) period because of the very bright diurnal periodicity. Of course, I could be wrong, but I haven't noticed a vivid weekly periodicity yet, so there's no way to identify it in my code.
I don't know, the timing logic should seem to give weekly periods. If you will find a daily pattern, then from them to identify relative to the days of the month or week periodicity is a matter of technique.
There are strategies, trend strategies that earn well at 40% accuracy, but the standard methods of MO do not allow to train them, dumping the class "1" to zero if the accuracy is not enough, and I just need such splits to separate and improve, so I'm looking for such methods. Otherwise Recall very small at 1 gets.
I should write custom loss fiures for these purposes
I do not know, the logic of the working time should seem to give weekly periods. If there will be found daily patterns, then from them to identify relative to the days of the month or week periodicities is a matter of technique.
I'm not saying that they do not exist. Most likely they are there, but this method is probably too coarse for their detection and isolation on the background of shorter periods and non-stationarity.
That's in theory... but in practice, no matter how you twist the glasses... )
I normalized for the volatility of increments, equalized the variance. It was only a loss of information.
I would not call it loss of information, but getting rid of misinformation.)
But it's not exact)
If trend, the TP is big and the SL is small. For example 500 to 100. Then at an error of 80%, there will be 20% of successful and 80% of losing trades. The balance will be near zero. If you trade leaves with an error of 70%, you will already be in profit. And if you find 50/50, the profit will be just enormous.
That's what I'm saying, but the completeness will be small, i.e. 100 trades out of potential 1000 will be traded (found class "1") of which 50 are unprofitable.
What does it mean to dump? 70% of errors it only seems to be dumped to class "0", the remaining 30% of class "1" can already make money.
30% - is a good result, but it is not always - now I'm writing an article, there is a strategy by MA, it turns out on average 5% detect units, well, it's almost on the standard oscillators with default settings :)
Since 2019 the sample is out of training. It doesn't leak - and that's good :)
The python one is a cpp wrapper. Everything works fine.
I mean it can be saved both in python format and in cpp. I save it in cpp and then convert it to mql by simple actions, since model itself is several arrays.There's a file saved in py format, it has a slightly different structure than CPP.
Yes, it turned out to be my mistake - I changed the concept of EAs and put the cart after the mare - it took me half a day to figure it out.
We need to write custom lossy functions for this purpose
And who knows how to train on such functions? The CatBoost on the pistol can count them and brake the training on them, but the calculation itself he does by his two.
grail found ))
I wonder how many of these grails are bought in the marketplace...
I wonder how many of these grails are bought in the marketplace...
earning, none))