Machine learning in trading: theory, models, practice and algo-trading - page 1649

 
Reg Konow:

The suggestion that we "read" Aristotle, Plato, or Hegel is ridiculous. They are not "read," they are studied. And if you don't know, Aristotle lived two millennia before Hegel. He did not and could not have a formed dialectical concept.

Plato is our friend, but the truth is dearer.

 
Evgeny Dyuka:

I trained it for an hour. Backtests show it trained better than 15 minutes. How in practice will be seen when the story piles up. Now I have a drawing for the night.

This is BTCUSD on M2. Network polling every 2 minutes. Every point has a forecast for 60 minutes, but it is very relative. The network is simply asked the question "what will happen in an hour?" and it answers the expected magnitude of the price change. The upper line is the downward movement, the lower line is the upward movement, and the thickness of the line is the degree of confidence. I will send you the link to the Expert Advisor with such a prediction later.

You were asked to give a series and a signal for some adequate interval, a third of at least of what you taught, that is, for 100 minutes, and then let's see how it predicts. But you give me small pieces that may sometimes look cool even with a mouse.

 
Wizard2018:

You don't say. The market is such a task, in which the main thing is just to understand "what to look for", and the rest is "a matter of technique".

nice words, but here it's like the beads before the pigs, here are mostly fans of fitting to garbage, fancy algorithms, like CNN and LSTM

 
Kesha Rutov:

You were asked to give a series and a signal for some adequate interval, a third of at least of what you taught, that is, for 100 minutes, and then let's see how it predicts. I don't know how to use it, but I've got a good view on it.

How do you give? Put Expert, it has already accumulated a story for about 10K minutes, i.e. 10K individual forecasts. I do not have more.
If you need not in the graph but in absolute values, I can send you the log file.
 
Is it realistic to use NS (or other methods) to get a prediction of the value of the price (or the waving) a step ahead better than "tomorrow will be like today" (that is, better than the incremental rate)?
 
Kesha Rutov:

golden words, but here it's like the beads before the pigs, here are mostly fans of fitting to garbage, with fancy algorithms, such as CNN and LSTM

Some people even advertise JPrediction_v15.0. What kind of people they are. Right?
 
Mihail Marchukajtes:
And some of them even PR JPrediction_v15.0. What kind of people there are now. Right?

Put up

 
Vizard_:

lay out

So I took it and laid it out. There are some people here who criticize me, and call me with a small letter, and I will lay out the leading technology for you. Yep. !!!!.

Let me say, I absolutely did not touch the class Broin-Robinson-Reshets. However, I tried a lot in the SippleSelexion class, which is responsible for the choice of the received model. And those innovations, which significantly reduce training period (reducing the price of the model) and of course the resulting models are more adequate. Of the latest:

And the quality of the model is not so good. I think it's just a coincidence... No more.

 
Ha ha, look at that. Has anyone ever heard of it at all? Boredom's about to explode :-)
 
Mihail Marchukajtes:

So I took it and laid it out. There are some people here who criticize me, and call me with a small letter, and I will lay out the leading technology for you. Yep. !!!!.


Will you sell it?