Machine learning in trading: theory, models, practice and algo-trading - page 1483
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 never left them. It's necessary if the new data exceeds a familiar range of values.
This is the outliers. They vary for every chip (I take 1 % from the top and the bottom), the volumes, for example, from 1 to 2474, I find the top 1 % and they start from 2000, after that everything above 2000 I equate to 2000. I cut the price delta by 0,01000, etc. At new data I do trimming by the same levels that I obtained at the training.
I.e. there is no need to bring all the features to 1.0.
This is what the emissions are. They are different for every branch (I take 1% from above and below), the volumes, for example from 1 to 2474, I find the top 1% and start them at 2000, after that everything above 2000 is equal to 2000. I cut the price delta by 0,01000, etc. At the new data I do trimming by the same levels that were obtained during the training.
I.e. I don't have to bring all the features to 1.0.
if the features themselves are non-stationary, I mean. Any stationary fixture is garbage 50\50, as far as I'm concerned. And if you cut them (emissions), then you should also prohibit trade in those areas.
Question for all neural networkers and random foresters - how well do your trained models behave when you change brokers?
I have recently started to feel that the difference in data from different brokers, in the aspect of machine learning, for some reason grows, although in theory it should be the other way around.
And maybe it's in the pre-processing...
The question to all neural networkers and random foresters - how well do your trained models behave when you change brokers?
For me it seems that lately the difference in data from different brokers, in the aspect of machine learning, for some reason is growing, although in theory it should be the opposite.
I don't know if it is because of pre-processing.
What brokers? In RF brokers are only on the exchange with one and the same quote, it makes no sense to discuss other kitchens, there may be any quotes
What brokers? In Russia, brokers are only on the stock exchange with the same quote, it makes no sense to discuss the other kitchens, there can be any quotes
What percentage of users of the terminal of the site owner on which we are discussing work with brokers with the correct quotes?
It's not really a question for me) on forex, all quotes are correct, but all are different. Right in the sense that it is impossible to prove the opposite
If i'm not mistaken, i don't see any difference, but if i'm asking about adequacy i see no big difference, at least for the m15 timeframe in the tester. So what about small things.It's not really a question for me) on forex, all quotes are correct, but all are different. Right in the sense that it is hardly possible to prove the contrary.
When I look at the chart I don't see any differences, but if I look at the m15 timeframe in the Strategy Tester I'll see only minor ones. I will try to do some small things.And what percentage of users of the terminal from the owner of the site on which we lead the discussion work at brokers with correct quotations?
It is assumed that Forex is a decentralized market and each broker has its own quotation providers and in order to give traders adequate prices the ticks are filtered
If you want to give adequate prices to traders the filtering of ticks takes place - search through the forum , "filtering", "ticks", found it right away
https://www.mql5.com/ru/forum/102066/page9#comment_2968124
i.e. all brokers always have the right quotes, and your TS is looking at the wrong one ))))
It is assumed that Forex is a decentralized market and each broker has its own quotation providers and in order to give traders adequate prices the ticks are filtered
If you want to give adequate prices to traders the filtering of ticks takes place - search through the forum , "filtering", "ticks", found it right away
https://www.mql5.com/ru/forum/102066/page9#comment_2968124
i.e. all brokers always have the right quotes, and your TS is looking at the wrong one ))))
Correct meant the same - stock exchange, which are centralized, please read at least one level up, not just the last words - there it was said that decentralized quotes from DCs makes no sense to discuss.
And you, are you ready to discuss only the administrative and legal aspects and who looks where, or there are your own models of MO and you can answer in essence - do they react to the variability of quotes?
The correct ones I meant the same - stock quotes, which are centralized, please read at least one level up, not just the last words - it was said there that decentralized quotes from brokerage companies make no sense to discuss.
And you, are you ready to discuss only administrative and legal aspects and who looks where, or there are your own models of MO and you can answer in substance - whether they react to the volatility of quotes?
Would you like a gimmick:
I used to have inputs that were quote-independent, because like many of us I always took the difference of the data in the resulting window. When you take the difference in a small window, as a rule this difference is always the same and the indicator value itself does not play a special role, but once I made a mistake when making an entry, which showed much better results than the ordinary difference, but the entries themselves became very sensitive to quotes. Since I train and prepare the TS on my home computer, and the robot is trading on the UPU, when I transferred the TS I began to notice that the signals are very different and the value fed into the network are also different from those that were in the training. Well, say two weeks ago the broker has changed the volume of one bar by one unit, as an example. And now the point.
Take AD calculated from a certain date. It's worth to change the volume of any bar at the beginning of the calculation, say 1 or 2 days after the start of the calculation. Just one bar and one change in volume and the result at the very end, ie at the current time will be different in numbers quite significantly. And that's where I learned the trick, how the broker throws the algorobots into the ditch. If you trade manually and with the help of the chart analysis, then this change will not be noticeable, but for a robot it will be very important. Now I test all parameters in full. Or do quote-independent entries....