Machine learning in trading: theory, models, practice and algo-trading - page 3147

 

maybe feel one of these?

Neural Oblivious Decision Ensembles

 
Maxim Dmitrievsky #:
Don't much like discussing someone's homebrews that aren't fully described.

especially if they don't work...

and all explanations are sermons in circles that don't say anything.

 
Renat Akhtyamov #:

Sanych, the 1500 bar H1 window is the same thing all the time: a tangent to the middle of the window.

It is said above - we go round and round.

That's right!

1500 is the statistic result for RF. Stability by classification error starts at about 1200 bar. Turned out to be true for other models as well.

 
Uladzimir Izerski #:

Balaboly are those who write articles and advertise themselves here on the forum, not trading ).

Only those who trade with unprofitable Expert Advisors and have a peacock's tail on the forum are worse than that.

 
mytarmailS #:

especially if they're not working...

and all the explanations are sermons in circles that don't say anything.

Everything I write here is Rattle or Caret, with some working refinements, but as systems knowledge in MoD it is these packages. Exactly from here: data exploration, preprocessing, modelling and estimation. And an error at any point brings the whole effort into the bin. The classification error I get less than 20% is the result of some improvements at each step of the described stages and within the stages. And an error at any point leads all efforts to the basket. It takes me a lot of time to collect statistics to justify this or that decision.

Unfortunately, on this branch there are only a few people with system knowledge in the MoD, which (knowledge) can put into the appropriate professional tools. And in this very specific criterion - fluency in R.

 
The snake bit itself on the R again.
There's not enough new blood in the thread. Remove the trolls and lure in a few real geeks, then it would be interesting :)
 
mytarmailS #:

especially if they're not working...

and all explanations are sermons in circles that don't say anything.

Well, if they don't work, it's also an experience :) but when it's not clear what's being discussed and leading questions don't help....
 
СанСаныч Фоменко #:

1500 is the statistic result for RF. Stability by classification error starts at about 1200 bar. Turned out to be true for other models as well.

62.5 days.

There's something to that, I'll admit.

because that's just over three months.

the average endurance of the average TC ;)

 
Maxim Dmitrievsky #:
Well, if they don't work, that's experience too :) but when it's not clear what's being discussed and leading questions don't help....

it's because the answer to a leading question can not be given in principle ))


One claims that catching rebounds is easy, you say to him - well, show him your trading, and he says to you: look at my orders on freelance ))))).

The second one claims about some mega grail, you tell him - how can you prove it? and he tells you: here's a picture of the Masonic triangle )))).

The third ....


It's not even funny, it's just some kind of dope in their heads...

 

I read quant.stackexchange, where quants don't believe in levels at all.

Here's one of the discussions.

Я голосую за то, чтобы закрыть этот вопрос как не относящийся к теме,
 потому что анализ графиков (поддержка и сопротивление)
 и стохастические финансы плохо сочетаются друг с другом ... 
– 
Ричи В
 15 февраля 2016 г., 8:41

That's how smart people deny the obvious...