Machine learning in trading: theory, models, practice and algo-trading - page 1561

 
elibrarius:
Tried a similar thing with TP/SL training signal markup. Even 90% of successful examples on the training section, give 50/50 on new data. I.e., it earns nothing.
However, I have to try it again, maybe I made a mistake somewhere.

But so far my opinion is that there are no regularities in the price behavior and that the price is the SB.

https://smart-lab.ru/blog/321245.php

Отдаю грааль в добрые руки
Отдаю грааль в добрые руки
  • smart-lab.ru
Картинка для затравки, Out of Sample perfomance (то есть результаты получены на выборке на которой не производилось построение модели): Сбербанк MOEX 15106 трейдов прибыль 140 руб на одну акцию, периодичность трейдов где-то 15 минут, equity где-то за год. То же самое для Ri, 22237 трейдов прибыль 100000 пунктов на один контракт, или 5 пунктов...
 

It follows from the conclusion:

Of course you cannot trade this because of the profits eaten by the commission, but maybe someone will be interested as an example of a working low-frequency approach.

I have used my TP/SL with allowance for commission and spread of a bar. This is why I haven't obtained such beautiful images of the balance.

 
elibrarius:

It follows from the conclusion:

I was placing TP/SL, taking into account the commission and the spread of the bar. That's why I didn't get such nice pictures of the balance.

I didn't even get these without taking markups into account )

 
 

Who knows what's wrong, why when teaching with markups turned on, when testing bad results. Somewhere an error in the code.

Z.I. No errors, the problem is in the model thresholds, I need to put higher that would overcome the spread. I will record a video later.

 

Hi guys,

I see you're loafing again, well, then keep the announcement.

Yesterday I set the broadcast time for December 4, 19:00 Moscow time, but you come early in 15-20 minutes I will have a preface for the locals.....

 
Mihail Marchukajtes:

Hi guys,

I see you've been slacking again, so here's the announcement.

Yesterday I set the broadcast for December 4, 19:00 Moscow time, but you can come earlier, 15-20 minutes before the broadcast I will have a preface for locals.....

Waiting. Do not forget to invite again closer to that date.

 
Maxim Dmitrievsky:

Who knows what's wrong, why when teaching with markups turned on, when testing bad results. Somewhere an error in the code.

Z.I. No errors, the problem is in the model thresholds, I need to put higher that would overcome the spread. I will record a video later.

Maybe it's not an error, the error is when you get a"grral" from one row with a couple of tens of thousands of candlesticks and some hour ones or even higher.

 
Kesha Rutov:

Maybe it's just not a mistake, but a mistake is when you get a "grrrl" from a shitload of one row of a couple of tens of thousands of candles, also some sentinels and higher.

Kesha, a mistake is everything that's connected with your attempts to think, another cognitive self-destroy, according to tradition, and there's no error detected here
 
Maxim Dmitrievsky:
Kesha, an error is everything to do with your attempts to think, another cognitive self-destroy, according to tradition, and no errors are found here

Well, mold it...