Machine learning in trading: theory, models, practice and algo-trading - page 1256

New comment
 
elibrarius:
By the way, I've got a situation where first separation almost doesn't improve error and second separation improves it by 100%.

I have 4 sectors with 10 points in each. 1 split, either along the x-axis or the y-axis. It will not almost improve the error, it will remain about 50%. For example, the first division is in the middle vertically. The second split in the middle horizontally will lead to a very strong improvement in error (from 50% to zero).
But this is an artificially created situation, it does not happen in life.

you can use a kernel (transormat data) and do it through one division. I don't know what kind of kernel for this case, but it definitely should be

Time series are not predicted like that, you need to distinguish cycles and periodic components. And since in the market, as the sample increases, these disappear, so the error of all 50/50

That's why only forecasting a few steps ahead works. With good regularization you get bigger cycles and the system lives longer, but there are less deals

 
Maxim Dmitrievsky:
Time series are not predicted that way at all, it is necessary to distinguish cycles, periodic components. And while in the market these cycles disappear as the sample grows, that's why everybody has 50/50 error.

I can't argue with that.)

 
Maxim Dmitrievsky:

That's why everyone has a 50-50 error.

Not everyone:) I have 10-15% error.

 
Kesha Rutov:

Not everyone:) I have a 10-15% error.

Me too, but it doesn't mean much on the new data... well better than 50 yes

 
Maxim Dmitrievsky:

Me too, but it doesn't mean much on new data... well, better than 50 yes

It's ok with new data, the problem is different, MO does not play any role in trading like indicators do, success depends on something else escaping formal interpretation.

 
Maxim Dmitrievsky:

Me too, but it doesn't mean much on the new data... well, better than 50 yes

If your profit is larger than your loss, 50 is the best.) You don't have to chase it, it's more than enough.
 
Kesha Rutov:

All is fine on new data, the problem is different, MO does not play any role in trading, just like indicators, success depends on something else, eluding formal interpretation.

That's right. Well, at least in general formulate this something, and then teach mo). With the clarification of mo itself can cope.
I already wrote: first the basic strategy, and only then the MO.
 
Kesha Rutov:

Everything is fine with new data, the problem is different, MO does not play any role in trading, like indicators, success depends on something else that escapes formal interpretation.

Everything has an effect on everything, right down to the day of birth.

 
Kesha Rutov:

Everything is fine on new data, the problem is different, MO does not play any role in trading, just like indicators, success depends on something else that escapes formal interpretation.

success depends on luck accompanying fools, only they can do MO and make it promising;)
 
Yuriy Asaulenko:
If your profit is greater than your loss, 50 is the best.) There is no need to chase after it, it's more than enough.

Yes, but it's not that simple, for example a random entry and TP/SL = 2 in the end is the same plum on the spread, because the stop will be twice as often as the profit, the market so easily will not overpower, so Soros and Buffett are quite rare.

Yuriy Asaulenko:
Exactly. Well, at least formulate something in general terms, and then teach your mo). Mo will be able to specify it.
I already wrote: first the basic strategy, and only then the mo.

And you know this "something", this "basic strategy (BB)"?

Maxim Dmitrievsky:

the totality of everything in the world influences ) right down to what day you were born

This is wisdom, omniscience, but how to find it without dying?

1...124912501251125212531254125512561257125812591260126112621263...3605
New comment