Machine learning in trading: theory, models, practice and algo-trading - page 1391

 
Grail:

All forwards, I don't even look at trader, Eurobucks, about a year

Approximately the same would happen on eurobucks and without MO if you were buying at the beginning of the week and selling at the end of the week during the year,

and then, like everyone else, would have safely merged, and how was it, in fact?

 

Every week I retrain them, then I carefully plug them in to make them work. These models go into production

In fact, this is redundant, because I will retrain them in a week anyway.)

This one is still on hold


This is for fx, for Forts is a little different, with adjustments for sessions, clearing and other stuff
 
Maxim Dmitrievsky:


It's clear that the model is being trained without any charts. It is better to tell me about the model itself.
 
Maxim Dmitrievsky:

Half of the topic is devoted to stories about my models, I think )) stopped writing already

I have not seen anything intelligible, or missed something. Well, no, so no.
 
Yuriy Asaulenko:
I have not seen anything intelligible, or missed something. Well, no, so no.

just forgotten already.

everything I write, every sentence is meaningful.

I'm not just arguing with everyone, I'm writing from experience

above supplemented, pre-post.
 
Maxim Dmitrievsky:

just forgotten already.

everything I write, every sentence is meaningful.

I'm not just arguing with everyone, I'm writing from experience.

above supplemented, pre-post.
I have not read the articles. I will read it.
In vain you put your trust in the MT tester as the ultimate truth. Maybe that's why the model lasts a week. It is strange.
 
Yuriy Asaulenko:
I have not read articles. I will read them.
In vain you put your trust in the MT tester as the ultimate truth. Maybe that's why the model lasts a week. That's odd.

Not only a week, but the efficiency decreases over time, which is natural

i'm testing with real ticks at the broker that i'm trading with so it's not a hope but accurate tests. although ticks are also redundant because they're not hft but to be sure how stops work, for example floating spreads, delays

 
Maxim Dmitrievsky:

I am testing with real ticks of the broker I trade with, so it is not a hope, but accurate tests.

i'm testing with real ticks of the broker i'm trading with, that's why it's not a hope but an accurate test. although ticks are also redundant because they are not hft, but to be sure how stops work, for example floating spreads, delays

For me 1m was always enough for tests. In fact, the ticks do not change much in the test, the tedium is larger. But in real market the ticks integrate into the system because they are needed only on the last candle and nowhere else. They go as Klose, and that's it.
 
Yuriy Asaulenko:
For my tests 1m was always enough. Ticks do not essentially change anything in tests, the tedium is larger. In real trading the ticks fit in the system because they are needed only on the last candle and nowhere else. They go as Klose, and that's it.

Of course, more if you have your own tester, in mt5 there is no drag

One sadness now - sockets do not work in the mt5 tester. I wanted to test advanced models on python, but I can't yet.

I'm not even going to bother with DLL, because I want to put the model on a remote server and connect terminals to it remotely
 
Maxim Dmitrievsky:

Of course, more if you have your own tester, in mt5 there is no drag

One sadness now - sockets do not work in the mt5 tester. I wanted to test advanced models on python, but I can't yet.

I'm not even going to bother with DLL, because I want to write the model on a remote server and connect terminals remotely
I think that sockets with python tester won't work correctly. At least because of thread inconsistency. And synchronize - well, I do not know, I can not say right away.
Reason: