Machine learning in trading: theory, models, practice and algo-trading - page 2209
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
Hi. Yes, that's right. The same markup is used in the tester. It's probably better to ask about articles in the articles. So that in one place was
I analyze feedback and see what can be improvedah ok))) just work work))) + the little one is growing up with little time))))
The question has matured. Some time ago I wrote a nouveau blog post on the topic of OOS location...
However, recent mutilations have made me think again about the following situation.
Let's say there are five years of randoms (no patterns). And after them two years of such, that stick almost any stick - the profit will grow.
I do training on the first five years. I get a straight line to the northeast on them. I start on OOS (two years after that) - of course, there is the same straight line.
Glad to find a pattern is complete nonsense in this situation. How do you not get fooled in such a way?
Obviously, this is quite a kindergarten issue in MO. Therefore must be dealt with 100% up and down. Who is in the subject, please share.
The question has matured. Once wrote a nubian blog post on the topic of OOS placement...
However, the recent atrocities have made me think again about this situation.
Let's say there are five years of randoms (no patterns). And after them two years of such, that stick almost any stick - the profit will grow.
I do my training on the first five years. I get a straight line to the northeast on them. I start on OOS (two years after that) - of course, there is the same straight line.
Glad to find a pattern is complete nonsense in this situation. How do you not get fooled in such a way?
Obviously, this is quite a kindergarten issue in MO. Therefore must be dealt with 100% up and down. Who is in the subject, please share.
I can only voice my opinion. The row is composed of contractions / classes, classes where there are profits, classes where there are not (well, very roughly if.) And if we define them, then all is well. If not, it's random. If we randomly formed classes that are on those 2 years, that can also happen, then it's fine, if not, it's random.
If there are classes that happen to be on those 2 years, that can happen too, then it's okay.
So the example is tailored so that everything is fruitful on the 2 years, whatever is "in" the randome.
This is not a hypothetical situation really.
It seems that the problem is reduced to automatic search for intervals, in which any stick is spiked. And from the OOS to throw out such intervals.So the example is tailored so that at 2 years, everything is fruitful, whatever is "in" the randome.
This is not a hypothetical situation really.
In '14 also worked well) The row for a long time was strictly defined).
You can vice versa, opt, teach, train on 2 fruitful years, and test on 5 random. If the test has a random result, then your classification is correct. 5 years random, 2 years with patterns on which the TS works with any setting.
It looks like the problem is reduced to automatic search of intervals, on which any stick is spiked. And then to drop such intervals from OOS.I do not think it is possible to reliably determine the beginning of a fruitful area for any TC settings in real life.
I think classification / clustering is our everything. And the speed of determining the beginning of a stable plot.
I do not think it is possible to reliably determine the beginning of a fruitful area for any TC settings in real life.
We are not talking about the real. It is necessary to find such areas in the history.
We are not talking about the real world. It is necessary to find such areas on the history.
Run the TS with different settings on the history for the entire period and find where all the TS are in the plus. Using them, determine how they differ from the others and how they are similar to each other.
Run the TS with different settings on the history for the entire period and find where all the TS are in the plus. According to them to determine how they differ from the others, and are the same with each other.
The problem is to determine this set. Then, of course, we find the power of the profitable subset of settings. And if it is large relative to the initial set, we have found it.
But the initial set to determine - it must be some kind of genetics. Not in the subject, in general.
The question has matured. Some time ago I wrote a nouveau blog post on the topic of OOS location...
However, recent mutilations have made me think again about the following situation.
Let's say there are five years of randoms (no patterns). And after them two years of such, that stick almost any stick - the profit will grow.
I do training on the first five years. I get a straight line to the northeast on them. I start on OOS (two years after that) - of course, there is the same straight line.
Glad to find a pattern is complete nonsense in this situation. How do you not get fooled in such a way?
Obviously, this is quite a kindergarten issue in MO. Therefore must be dealt with 100% up and down. Who is in the subject, please share.