Machine learning in trading: theory, models, practice and algo-trading - page 1254

 
Maxim Dmitrievsky:

In general, I've looked through a lot of theory about approximate trajectories, and a lot of math... I'll have to study it for a while. So far I see the potential in terms of iterative transformation of traits (i.e. you can add a degree of interpolation at each iteration and the trait will change), but there are some other tricks that I have not yet understood

https://arxiv.org/pdf/1603.03788.pdf here is a general use for machine learning

https://arxiv.org/pdf/1307.7244.pdf and here is piecewise linear interpolation for quotes and something else

I don't know, but it may be relevant only if you want to make a fully self-learning TS, the high level, so to speak.
Did you learn how to do preprocessing by these articles or by other ideas?
 
elibrarius:
Well, the bank of forex is so big that one can't eat it alone)))) At least because it includes all the other banks in the world.
Maxim, I told you - write, do not write - it is not perceived. Information is rejected. They are immune.
 
elibrarius:
Did you learn how to do preprocessing from these articles or from other ideas?

No, I haven't finished that yet, I've been practicing on python, made a link to mt5 via sockets, but I haven't trained the models yet.

I cannot understand why the results will improve so much... if i finish this, i may draw an article, but if it will be a real gem, i won't do it, i will trade it by myself.

Who would read them? ))
 
Yuriy Asaulenko:
Maxim, I told you - do not write, do not write - is not perceived. The information is rejected. They are immune.

Yuri, are you drunk? ) I did not write that.

But in general I agree in part, but it does not matter as the market cannot be beaten completely all the same

 
Maxim Dmitrievsky:

Yuri, are you drunk? ) I did not write that.

But in general I agree in part, but it's not very important because you can't beat the market completely all the same.

You didn't write it, but I addressed to you. Or you're allowed only when you're drunk? )) (Just kidding.
It amazes me that they all trade on some global intergalactic market. It makes no sense to explain anything.
But it's no use trading with such ideas.
 
Maxim Dmitrievsky:

No, I haven't finished this yet, I've been working on python, made a link with mt5 via sockets, but I haven't trained the models yet.

If I will finish this, maybe I will write an article, if it will be a real "grail", I won't do it, I'll trade by myself.

Who would read them?)
My 7-10% error has worked only with gameplay (a year ago due to inexperience). Until I removed the peeking predictor.
 
elibrarius:
I only got 7-10% with a peek into the future. Until I removed the peek predictor.

I don't know how to do peeking, because I have never used zigzags and so on.

entropy or rms itself can be 0.5, I'm talking about misclassification

 

And that's the thing... nothing special can be achieved with raw quotations of the 1st instrument as predictors (we mean any transformations as well), well, almost everyone who is more or less familiar with MO has already done that. It only remains to adjust the model to the market situation and that's all

i.e. take simple momentum strategies or mean reversion, and apply the models to markets at some stage. There will be a lot of handjob all the same

 
Maxim Dmitrievsky:

And that's the thing... nothing special can be achieved using raw quotations of the 1st instrument as predictors (any transformations are also meant), well, almost everyone who is more or less familiar with MO has already done that. It only remains to adjust the model to the market situation and that's all

i.e. take simple momentum strategies or mean reversion, and apply the models to markets at some stage. There will be a lot of handjobs all the same.

It won't work with simple ones either. There will be not enough information for MO. Or on the contrary - too much - it will require very complex MoD, which is unaffordable.

 
Yuriy Asaulenko:

It won't work with simple ones either. The information will be insufficient for the MO.

If the market is more or less stable, a trend or something, then this time will work, at least for me... the patterns are the same, why not?

I don't know if it works, but I've simplified it to one button pressing and no predictors )) Such a funny looking machine that I may sell it as an exhibit of human madness