Machine learning in trading: theory, models, practice and algo-trading - page 3634
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
1. Already shared. You have all the possibilities and more.
2. To whom as you like, you don't want as you like.
3. Above gave the post, if not useful, you lose in the contest. If useful, you have a fighting chance.
4. Why. Combat verification. Otherwise it's all water and bollocks.
5. I do not think that everyone is stupid, on the contrary, I challenge. This branch is read by hundreds of thousands of people, so the risk is high, I can lose.)))))))) Only positive, bodybuilding, health, victory over the market, all business. The main thing is not winning, the main thing is participation. All those who refused to participate - he is not confident in himself, in his methods and approaches. A weakling in general. Just do not need to say, like, here, my methods are correct but I can not participate, piss, or afraid, or not sure. Either you can and solve problems, or you can't do a thing and are a simple fuflomet. The third option is to listen to the former.
6. You can, but you haven't suggested anything yet, and I can suggest it right now.
Peace be upon the world, MOU MO, Neyru Nuiro, etc.
1. It seems that you are very worried that I was told something so valuable that I might have some competitive advantage. I must reassure you that I was not able to identify any new and unique information for myself. Perhaps I'm a feeble mind. So relax - I will not reread it.
4. So on what data do you propose to build a model now? I don't understand anything.
1. it seems that you are very worried that I have been told something so valuable that I may have some competitive advantage. I must reassure you that I was not able to identify any new and unique information. Perhaps I'm a feeble mind. So relax - I won't reread it.
4. So on what data do you propose to build a model now? I don't understand anything.
1. Too bad you think you have a feeble mind. And sorry I thought otherwise, I hope I'm wrong now.... Anyway, I am confused as to what you are confused about.
2. If I say in advance what data the participants will base the model on, the intrigue is dispelled, right? - I told you, the formula is very simple, but that's the hocus-pocus, that even a simple formula will be a serious problem for MOSHniki, not like a real non-stationary market, the formula of which is not known at all and it is not known if it has a formula at all.
What is highlighted in bold does not make much sense, but it sounds beautiful. The market is like that...))))))
Tomorrow I will post here a part of the graph of an example of such a function, and let's see who can guess how the function will look like at the same length of samples in the future. And then after a while I'll post the formula for that graph..... Just manually guessing what the graph will look like on the oos.
"In reality, it's completely different than it really is." - or so said Comrade Exupery.1. Sorry you think you possess a feeble mind. And sorry I thought otherwise, I hope I'm wrong now.... Anyway, I am confused as to what you are confused about.
Well, I was once told that my intelligence does not correlate with wealth in the form of material assets, and therefore knowledge is useless or something....
2. If I say in advance on what data the participants will base the model, the intrigue will be dispelled, right? - I told you, the formula is very simple, but that's the hocus-pocus, that even a simple formula will be a serious problem for MOSHniki, not like a real non-stationary market, whose formula is not known at all and it is not known if it has a formula at all.
What is highlighted in bold does not make much sense, but it sounds beautiful. The market is like that...))))))
Tomorrow I will post here a part of the graph of an example of such a function, and let's see who can guess how the function will look like at the same length of samples in the future. And then after a while I'll post the formula for that graph..... Just manually guessing what the graph will look like on the oos.
"In reality, it's completely different than it really is." - or so said Comrade Exupery.Ah, got it, there's no public data yet.
However, guessing from pictures is not for this thread. Functions can be very intricate, well, even take vector graphics, you know about parameterisation of curves....
No one here claims that you can reconstruct the formula of any function by a piece of points.... or has a forum member directly stated this to you?
On the contrary, if a function behaves differently with new values of variables, then we should talk about unrepresentativeness of the sample and complications in training.
I just don't understand why you should prove something that nobody argues with?No one here claims that you can reconstruct the formula of any function from a piece of points.... or did a forum member explicitly state it to you?
That's not what we're talking about. Networks do not reconstruct the formula of a process, but approximate it. But the point is that the same form of the approximation result can be generated by different sets of weights, and these different sets will have the same metrics (e.g. errors). And only one set (with slight variations) out of many possible sets "correctly" describes the process being approximated. Many people, and you among them, argue that the fitness function is not important (or set of metrics), but if the ff does not adequately estimate the process, it is impossible to get not only a robust set of weights, but there will be no reproducibility of results at all.
Here is an illustrative example of the process:
and here is the same process, but at a different site t:
even further away:
Is the process non-stationary? - No, it's stationary, the formula is the same, here it is:
F := sin(x)/4 + cos(x^2)/4 + 1/2
Only some combination of weights from all possible will allow the network to go through all these parts of the process (no matter how long it takes), although outwardly the process looks like non-stationary. The market is an analogy.
All sorts of artificial tricks, like "undertraining a little bit" and others, will accomplish nothing if you don't use proper judgement in training. If the autopilot navigator is set up incorrectly, the car will never arrive where it needs to go.
So the robust set of neuronics is nothing but a reproduced formula of some process (regularity), just written in the form of relations instead of analytically. Accordingly, if the formula is not reproduced accurately, it will not work in new areas (it will gradually deviate from the actual process, but more often it will deviate sharply).But the point is that the same form of the approximation result can be generated by different sets of weights, and these different sets will have the same metrics (e.g., errors).
This is possible.
And only one set (with small variations) out of many possible sets "correctly" describes the process, approximating it.
What do you mean "correctly", with less error on new data or no error at all?
Can you prove that only one and not more?
Many people, you included, argue that the fitness function is not important (or set of metrics)
I was arguing that standard metrics do not guarantee the best split for a model in terms of its robustness on new data. I was talking about tree models. I haven't changed my mind yet. Or what do you mean?
but if the FF does not adequately estimate the process, it is impossible to get not only a robust set of weights, but also will not be reproducible results at all.
Which FF do you propose to use in this case for NS training? How do you evaluate its "suitability"?
So the robust set of neuronics is nothing but a reproduced formula of some process (regularity), just written in the form of links and not analytically. Accordingly, if the formula is not exactly reproduced, it will not work in new areas (it will gradually deviate from the actual process, but more often it will deviate sharply).
I can see that you have better understood the essence of the learning process, and this knowledge has given direct inspiration! This is very good!
Very difficult to predict
It is sad that for all the time of "communication", the optimiser has not shown anything useful for trading and MO. After all, a misguided person can only mislead.
You are a strange passenger - always whining - regularly someone has not shown you something, has not given you something.
back in 1987, there was a film with Stallone called "To the Best of Your Ability" -- the key phrase there: "Nobody in the world is going to meet your needs. If you want something, take it yourself."