Machine learning in trading: theory, models, practice and algo-trading - page 1627
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
Well, let's say not all minutes, but those with a body greater than N-points as an example
The idea is correct in theory, but in practice it's all bad. It is supposed that "informed investors" (institutors, insiders, etc.) should obviously behave differently in the market, for example, they should yap at the market with huge orders, put unbreakable "plates" in the market, cancel with a series of equal orders at equal time intervals, etc.. Then this behavior can be recognized and exploited. Maybe it was like that some time ago, in part, but not now. Not to mention OTC and darkpools where large volumes are mostly traded, all institutions and close to them have long been very qualitatively disguised as noise. Of course it is impossible to hide the throwing in of large liquidity, but it is done very visually and differently each time. So filtering nowadays just reduces the data, increasing the probability of fitting.
And also Enokenty, I demand from you a public apology for what you equated me, such a worthless programmer and handyman to an outstanding personality such as Reshetov Yury. If you saw his code and the manner of his writing, you would have admired it as I admire the manner of its programming. Yes, I made adjustments to the optimizer, which, to me, have improved the final outcome, but to compare it to me is stupid. Compared to him, I'm just a prep school student who always skipped classes and is always shouting from the back of the school "What? So I'm waiting for an apology.
I do not know, it's a question of who to apologize to, "Misha" or Jura, especially after this self-deprecating and again Reshetov-praising post)))
Reshetov is nothing to read as well as me or you, this choice of an idol is very strange (as for a non-clone). If you were a fan of Misha Malyshev or Jim Simons, I would understand it, but Reshetov is just a forum chatterbox and mediocre algotrader, only his clone can be a fan of him, the probability of it is close to 100%. To make a foolish clone, I think, is not difficult, when there is nothing to do, but you strongly stumble with Reshetov's advertising, it is very obvious.
The idea is correct, in theory, but in practice all is bad. This assumes that "informed investors" (institutionalists, insiders, etc.) must obviously behave differently in the market, for example, they must plow through the market with huge orders, put impenetrable "plates" in the market, or throw a series of identical orders at equal intervals, etc.. Then this behavior can be recognized and exploited. Maybe it was like that some time ago, in part, but not now. Not to mention OTC and darkpools where large volumes are mostly traded, all institutions and close to them have long been very well disguised as noise. Of course it is impossible to hide the throwing in of large liquidity, but it is done very visually and differently each time. So filtering nowadays just reduces the data, increasing the likelihood of fitting.
I've read his article for the second time, the first time I read it a long time ago, I didn't understand anything... If I don't understand something, I'll try to summarize his approach, let him correct me...
Mykhi has a signal of the Sequent trading system, but in fact it can be anything - a crossing of wands, a black candle at 10 am, etc. He says it can be anything.
In human terms: we simply subjectively reduce the dimensionality in the data, reducing the degrees of freedom, for AMO (machine learning algorithm). And this is probably correct.
2) further myha takes into account "market context" reports, open interest and considers a signal from the sequent system in the context of the "market context" (sorry for the pun) and trains a network for each such combination...
In human terms: The "market context" is also essentially a cluster, and it can be anything. In a situation where one cluster (point 1) is a nested cluster (point 2), we reduce the dimensionality even further, by orders of magnitude. And now we train AMO on received compressed data. And this is probably right too.
3)We teach AMO1 classification to three classes "bye", "sat", "don't know" (Michael taught it to "bye" and "sat" separately, but I don't see the point).
4) on the "new data 1" we see the error of recognition AMO1 and create new tags in the classes of "bye/no guess" or "bye/guess" or "sel/don't know
5) Train the second AMO2 on the data and answers from the first AMO1
6) On "new data 2" watch the recognition error of AMO2
In human terms: With the second AMO2, we predict whether or not AMO1 will correctly guess its class
right mikha?? see I put your whole article in 10 lines))
In the case of substantial non-stationarity, it is more correct to speak of multifractality, since the fractal characteristics change over time. These changes are as unpredictable as any other.
I don't mean Hearst's and so on, I mean more simple thinking that the patterns (repeating patterns) definitely exist, they can't exist but they are always different in sizes and that's why systems don't work, AMO doesn't learn, coconut doesn't grow)
It displays neural network predictions by putting colored areas on the chart. There are alerts and flushes.
It is possible to change the threshold of network confidence - from 0 to 100+, at zero it will show everything, i.e. at every minute.
!!! It works only on BTCUSD and only on M1 timeframe!
It gets information from neuro via API, so you have to enable WebRequest in the terminal.
How to configure it is in the attached instructions.
Download here.
I do not know, it's a question of who to apologize to, "Misha" or Jura, especially after this self-deprecating and again Reshetov-praising post)))
Reshetov is nothing to read as well as me or you, this choice of an idol is very strange (as for a non-clone). If you were a fan of Misha Malyshev or Jim Simons, I would understand it, but Reshetov is just a forum chatterbox and mediocre algotrader, only his clone can be a fan of him, the probability of it is close to 100%. To make a foolish clone, I think, is not difficult when there is nothing to do, but you fall heavily with Reshetov advertising, it is very obvious.
I have completed my MT5 Expert Advisor.
It displays neural network predictions by adding colored areas to the chart. There are alerts and flushes.
I can change the confidence threshold from 0 to 100+, at zero it will show everything, i.e. at every minute.
!!! It works only on BTCUSD and only on M1 timeframe!
It gets information from neuro via API, so you have to enable WebRequest in the terminal.
How to configure it is in the attached instructions.
Download here.
I updated + history, I can see how neuro thinks. For me her vision of the market is striking, there is no direct connection to the indicators. Expert now here.
Not perfect, but something is thinking... Just overbought/oversold cannot be explained.