Machine learning in trading: theory, models, practice and algo-trading - page 1606
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
firstly, what other feilish regressors? what kind of nonsense, why then when you complicate the problem MGUA also goes out?
Secondly, my example has the same data for both the MGUA and the booster
thirdly, you don't need to mess around, can't you make a matrix with four random values in python and then make a cumulative sum of them? To check your boost?
2 lines of code ))
I'm wondering what the hell is going on there myself.
mgua creates fake variables from source, (depending on the kernel you use)
MGUA creates fake variables, (depending on the kernel you use)
Forget about MSUA, I'm telling you - create a similar dataset like mine and run your boosting on it and see what you get, WITHOUT MSUA, just forrest or whatever you like. Or send you a text file with my data?
Forget about MSUA, I'm telling you - create a similar dataset like mine and run your boosting on it and see what you get, WITHOUT MSUA, just forrest or whatever you like. Or should I send you the textbook with my data?
yes i'm telling you why the boosting is poorly trained ka cusumma you and the mgua is good. Because of fake regressors, polynomial regressors, for example.
take a linear regression x on y, add x^2, x^3 as fake regressors, you get a polynomial reg that fits the curve
and the forest won't fit as well on x alone. And mgua masses fake variables on an industrial scale
I'm talking about the technical part of the question. That's why you think that mgua is great and boosting is crap. Because you don't know how to useEugene Good day, thank you very much, at least for the fact that you are a practitioner and not another rubbishman of which there are 95%.... What you do(test on "third" sample) in terms of GMDH(method of group consideration of arguments) is called "criterion of predictive ability"http://www.machinelearning.ru/wiki/index.php?title=%D0%9C%D0%93%D0%A3%D0%90#.D0.9A.D1.80.D0.B8.D1.82.D0.B5.D1.80.D0.B8.D0.B9_absolute_noise-immune
Let's remind that the first publications about GMDH began somewhere since 1960-th those "your know-how" idea with the teston the "third" sample is already 60 years old)))
But I'll notice that the approach never gets old, so I strongly recommend to read worksof A.G. Ivakhnenko...
For example MSUA regression just mocks the regression of modern random forest algorithm and all sorts of boostings ...
Now about the links to Telegram... I found nothing but signals there, but it's interesting to read your approach and your way of thinking, Dmitry was right to say that it is necessary to publish here, though in an openly boorish form...
I'm simply stating that I was able to train a neural network and its signals in the real market confirm that it is adequately trained.
This is the first public demonstration of a working network amidst the general failure of this topic and lack of reliable results.
If you watched the signals, you must have noticed that the network reacts correctly to the market. Moreover, its behavior cannot be explained by our usual trading strategies or binding to the indicators, on the contrary, its behavior is often illogical.
The effectiveness at this stage does not matter, the main thing here is the fact that it is possible at all, and the quality of the prediction can be improved infinitely, it is a matter of time and equipment.
Yes I'm telling you why boosting is poorly trained as a cusumma for you, and the mgua is good. Because of fake regressors, polynomial regressors, for example
take a linear regression x on y, add x^2, x^3 as fake regressors, you get a polynomial reg that fits the curve
and the forest won't fit as well on x alone. And mgua masses fake variables on an industrial scale
I mean the technical part of the question. That's why you think that mgua is great, and boosting is crap. Because you don't know how to useYep... got it ))
But it still comes out good that fake regressors of MGUA
And it's bad that forrest is not producing fake regressors.
Because MSUA can handle the same data "out of the box" and boosting needs to create these regressors manually... And what to create them I don't know, it all depends on the data
I do not understand this subtle irony, what does the GMDH have to do with it? I didn't claim that this is my know-how, it's just a routine test of the results.
I'm simply stating that I was able to train a neural network and its signals in the real market confirm that it is adequately trained.
This is the first public demonstration of a working network amidst the general failure of this topic and lack of reliable results.
If you watched the signals, you must have noticed that the network reacts correctly to the market. Moreover, its behavior cannot be explained by our usual trading strategies or binding to the indicators, on the contrary, its behavior is often illogical.
Effectiveness, at this stage does not matter, the main thing here is the fact that it is possible, and the quality of the prediction can be improved infinitely, it is a matter of time and equipment.
Never mind, about the irony ))
The signals in the form of text messages, it's hard to compare the performance in the market, I would be happy to see the trade in a more visual form. And again, not a single word about the algorithm of actions to create a trading algorithm, what are the chips, what is the target, how the data is preprocessed, etc.
Max! Have you tried using associative rules to find patterns like the arriori algorithm or something like that, I like this approach
Forget the irony))
The signals in the form of text messages as it is hard to compare the performance in the market, I would be happy to see the trade in a more visual form. And again, not a single word about the algorithm of actions to create a trading algorithm, what kind of chips, what is the target, how the data is preprocessed, etc.
1) only tf M1, because the predictions are not tied to timeframes,
2) the indicator will need to request the information via sockets from my server, because it is unreal to run tensorflow from the client.
3) Now the calculation of all the models for each candlestick takes 12-13 seconds, it will take me more time, soon it will be overloaded by the equipment...
The second option - to try to make an indicator for tradingview, but it is not certain that pine supports web sockets. There are no other options, drawing charts retroactively - no one will believe it.
As for the algorithm and stuff - I'm ready to answer any questions except the logic of selecting the input data for training.
Yes, visualization is necessary. Signals are crooked. There is an idea to make an indicator of AO type - under each candlestick there is a bar with prediction force above and below zero. But there are problems:
1) only tf M1, because the predictions are not tied to timeframes,
2) the indicator will need to request the information via sockets from my server, because it is unreal to run tensorflow from the client.
3) Now the calculation of all the models for each candlestick takes 12-13 seconds, it will take me more time, soon it will be overloaded by the equipment...
The second option - to try to make an indicator for tradingview, but it is not certain that pine supports web sockets. There are no other options, drawing charts retroactively - no one will believe it.
As for the algorithm and stuff - I'm ready to answer any questions except the logic of selecting the input data for training.
Well, it's hard to ask anything, it all starts with preprocessing data, and that's what you do not want to talk about. (
Well... I'm interested in
1. Whether the algorithm works for currencies
2. Does it build the forecast for a fixed length of n forward candlesticks or the network itself will decide on the length
3. Why it takes so long to process the signal 12-13 sec per candle
4. Why do you want to broadcast deals publicly?
5. for the forecast, use the data in the form of a function (price, indicator) or something trickier.
the best visualization is deal
Yep... got it ))
But it still comes out good that fake MSUA regressors
And the fact that Forest does not produce fake regressors is bad.
Because MSUA can handle the same data "out of the box" and boosting needs to create these regressors manually... And what to create them I don't know, it all depends on the data
there are special separate libs for generating fictitious features, then they also go to listing, it will be the same
mgua algorithm is weak because it uses a simple regression, so it multiplies fics out of the box