Machine learning in trading: theory, models, practice and algo-trading - page 3001

 
fxsaber #:
Share tst files from TC on MO. Balance line is not important. The main thing is 10K+ non-overlapping positions in time.

Yesterday it turned out that most people do everything in R.

 
Happy anniversary, comrades!
I'm sure that even on the 6,000th page there will still be cactus gnawing here).
Forward, to new achievements!
Well, you can ban now)
 
Maxim Dmitrievsky #:

from 2000 to 2010 pours, from 2010 to current works.

didn't really understand what it did (nothing, I guess) and what to do with it further :)

default settings


It seems to cut out the unsuccessful time periods and leave the successful ones.

I did an experiment in my place - just 24 models, 1 for each hour. Among them there are randomly successful ones. But for almost all of them the first split on other chips worsened OOS. With the caveat that on one target with one set of features. With others it may be different.
If time is added, it is not forced, but if the model chooses it, then - ok.

 
Maxim Dmitrievsky #:

from 2000 to 2010 pours, from 2010 to current works.

trades on the hourlies

didn't really understand what he did (nothing, I guess) and what to do with it further :)

default settings


5 digit appeared in the 10th...

 
Forester #:

It sort of cuts out the unsuccessful periods in time and leaves the successful ones.

I did an experiment with myself - just 24 models, 1 for each hour. Among them there are randomly successful ones. But for almost all of them the first split on other chips worsened OOS. With the caveat that on one target with one set of features. With others it may be different.
If time is added, it is not forced, but if the model chooses it, then - ok.

Got it. The keys just don't work that way on a mac.

on that test on the training sample intervals were almost not thrown out, but on oos almost all of them were thrown out :)
 
Maxim Dmitrievsky #:

I figured it out. The keys just don't work that way on mac.

on that test on the training sample the intervals were almost not thrown out, but on oos almost all of them were thrown out :).

Unfortunately, missed the tst file.

 
fxsaber #:

Unfortunately I missed the tst file.

I'll show the others later, this one was boring.
 
3000 pages...

Most productive branch
 
Ivan Butko #:
3,000 pages...

Most productive thread.

Yes, this ward has the most difficult patients.... :)

 
Ivan Butko #:
3000 pages...

The most productive thread.
Just here people know how to test their algorithms, and not immediately after fitting they run to the market
Reason: