Machine learning in trading: theory, models, practice and algo-trading - page 2588
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
But there is no reaction from MetaQuotes ((.
And there won't be, because the real spread spoils the advertising image. At most they will offer to use custom symbols.
And it won't, because the real spread spoils the advertising picture. At most, they will offer to use custom symbols.
That's too bad. The simulation of spread in time would make testing closer to the real.
And it won't, because the real spread spoils the advertising picture. At most, they will offer to use custom symbols.
Advertising should say that our tester is the most accurate, compared to other platforms.
This is more likely to attract developers and users. Than the first tests and their comparison will show that the testing in MT5 is not true for the opening prices.
Advertising should say that our tester is the most accurate, compared to other platforms.
This is more likely to attract developers and users. Than the first tests and their comparison will show that testing in MT5 by opening prices is not true.
That's too bad actually. Modeling the spread by time would make the testing closer to the real.
By advertising I mean the data on the symbols on the brokerage company pages. It is CAs, not developers who are clients (bring money) for metaquotes. Free of charge for end users has its disadvantages.
By advertising, I mean data about symbols on CA pages. It is the DCs, not the developers, who are the clients (bring money) for the metaquotes. Free for the end user has disadvantages.
Lack of possible or expensive to implement, having the possibility in understandable conditions and prices, and free. Of course free is often skewed with negativity))) As well as the lack of possibility or the high cost of implementation))))
https://stats.stackexchange.com/questions/31513/new-revolutionary-way-of-data-mining
Yes, that's what he says. And commentators - well, they are datascientists, not traders. They with their stationary datasets can't understand our traders' problems.
Yes, that's what he says. And commentators - well, they are datascientists, not traders. They with their stationary datasets cannot understand our traders' problems.
As I understood it.
I explain it in the simplest way possible.
If you think of the TS as a pattern, in order to check whether it is a true pattern we should optimize its parameters and if the TS will almost always earn at different parameters, then the pattern (TS) is true
And how did you figure it out?
If you look deeper and from AMO's point of view
You need to visualize the search of parameters during training; if optimization surface looks like noise
this is most probably a local maximum in the noise and this is retraining and a nonworking model
You should strive for the following picture
Where the model has a well-defined "island" of working parameters, i.e. it is necessary to optimize its parameters and if the TS practically always will earn with different parameters, then the pattern (TS) is true
That's how I see it, I may be wrong, but...
How I understood it.
I will explain it specifically in the simplest way possible.
If you imagine TS as a pattern - in order to check whether it is a true pattern it is necessary to optimize its parameters, and if the TS will almost always earn at different parameters, then the pattern (TS) is true
And how did you understand it?
Well, something like that. Everyone understands that overfitting is evil and is looking for ways to protect against it. What the guy is saying is that the "test on oos" way isn't such a good tool to protect against overfitting. The simplest example, there are 10,000 people, each flip a coin 10 times. We selected the people who had all the flips that were eagles - oh, these guys know something. We asked each of them to flip 10 more times. Now, these three guys are some kind of losers, but these 3 guys got 8-9 times out of 10 again. Oooh, they can do something. Clearly, this situation is possible purely because of random rolls (not the same millions of monkeys who can write War and Peace by accident). It's the same with strategies. So you have to use other ways, or if oos, then wisely too. The alternative they offered is yes, something like: you better watch something like average. Like, if the coin is flat, it will fall evenly here and there in the average. But if it's heavier somewhere, then you'll see this skew by the average results of those 10,000 people. Somehow)).