Machine learning in trading: theory, models, practice and algo-trading - page 641
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
On Sensei, a fit from the boys))) h2o.automl.
The rattle is average, but it's all on automl...
Take it away bro!!!! I knew that everything was already stolen (made) before us (C) "Operation Y"
Is it necessary on forex data where patterns are hard to find? It seems to me that half of the examples may be eliminated by such a program. The outliers can be searched for with simpler methods: do not delete, but, for example, equate to the maximum allowed.
I tried to substitute 20000 lines of real data.
At EMVC_MIN_TRUST <- 0.9 it is suggested to delete 18488 lines
If EMVC_MIN_TRUST <- 0.5 - it is suggested to delete 8110 lines
I don't know how to use this filter in real trading. It would be good if the new data would pass through it and we could discard unsuitable examples, but since we don't know the answer, we will not be able to use them in this function.
I.e. I will have to feed the data to the NS without filtering. And if all such examples were screened out before training, then for the NS it will be an unfamiliar area of data, and it will give a random prediction.Is it necessary on forex data, where patterns are hard to find? It seems to me that half of the examples can be eliminated by such a program. And the outliers can be searched for with simpler methods: do not delete, but, for example, equate to the allowable maximum.
Finding patterns, imho, is not just difficult, but impossible. It is interesting that the eye is able to find them during manual trading. And it is not intuition but knowledge.
Attempts to formalize this knowledge turn out to be worse than hands, because it is a very multifactorial system. Consequently, formalization should be left to the methods of DM themselves, without interfering or imposing their vision of the process on them. However, it is possible to limit the field of functioning of DM. We come to systems which are combinations of ordinary automata with DM that complement these automata.
So what's the question about finding features?
there is only price in our case. Any price transformation is a priori regularities, in the form of some "memory" of the process (indicators plotted over n-periods). That is, if we do not know the regularities, we can only input price, increments with different periods to account for memory processes.
what can be other than price increments? or not, what are you picking up there so scrupulously, is there any? :)
There is an atvoregression process with order, you can do the same thing through NS. In my opinion, this is the only thing that can be taught. I mean take econometric models and extend them
IMHO... that's why I don't even try to pick up chips :) and nerves are fine (but not really)
in other words, what can we find in the price: trend, seasonality, cyclicality, noise
So what's the question about finding features?
there is only price in our case. Any price transformation is a priori regularities, in the form of a certain "memory" of the process (indicators plotted over n-periods). That is, if we do not know the regularities, we can only input price, increments with different periods to account for memory processes.
what can be other than price increments? or not, what are you picking up there so scrupulously, is there any? :)
There is an atvoregression process with order, you can do the same thing through NS. In my opinion, this is the only thing you can teach NS. I mean take econometric models and extend them
IMHO... that's why I don't even try to pick up chips :)
NS can be taught pattern recognition, sets. Consequently, we can formalize what we are able to do, and leave the rest to DM (NS, etc.). I.e. we have a ready autosystem on indicators-logic. And to fit unformalizable things into it we already use DM (NS etc.), cutting off all unnecessary things from DM methods in advance... The features, as you call them, will be found by DM (NS) itself.
I can only say that this approach works. At the same time the methods of DM themselves are significantly simplified.
The NS can be trained to recognize images, sets. Consequently, we can formalize what we are able to do, and leave the rest to the NS. I.e. we have a ready auto-system on indicators-logic. And to fit unformalizable things into it we already use DM (NS and others), cutting off all unnecessary things from DM methods in advance... The features, as you call them, will be found by DM (NS) itself.
I can only say that this approach works.
if we have TC , but if there is no TC or there are no clear rules? then you have to read what people wrote on a silver platter for thousands of pages )
this is wrong, ns will not find chips! chips are fed to the input... it's naive to think that the ns will find stable patterns from the set of crap given to it... the key word is stable
This is if there is a TS , and if there is no or no clear rules? then you need to read what people wrote on a silver platter for thousands of pages.)
) You can build a TS at least on two MAs. )
If there is no TS, then probably we need a very complex system of DM. If we have even a simple TS, we cut off from DM a lot of information that does not need to be processed by DM, i.e. we pre-filter the bazaar).
And this is the G... in the input stream we have significantly filtered.
The TC can be built on at least 2 MAs. )
If there is no TS, we probably need a very complex system of DM. Even if we have a simple TS, we cut off a lot of information from DM that does not need to be processed by DM, i.e. we pre-filter the bazaar).
And this is the G... ...in the input stream we've filtered out significantly.
so what's the point of feeding it in the first place? ) if apart from increments all the G...
there are non-periodic cycles in the market, each one is characterized by an internal "memory" with a certain lag
All econometric models are based on it, nothing better has been invented yet
So what's the point of submitting it in the first place? ) if besides increments all G...
There are non-periodic cycles in the market, each one is characterized by an internal "memory" with a certain lag
All econometric models are based on this, nothing better has been invented yet.
This does not apply to my systems). I work on short intervals. In general, I don't even care what happened yesterday.
I don't insist, but it's my belief that a long-term forecast, even for a few hours, is impossible in principle.
My systems are not concerned). I work on short intervals. I generally don't even care what happened yesterday.
I do not insist, but it is my belief that the long-term forecast even for a few hours is impossible in principle.
Look at my threads with predictions )
I do not need to do that, but it's pretty obvious and realistic for me