Machine learning in trading: theory, models, practice and algo-trading - page 1303

 
Maxim Dmitrievsky:

10% error per test and trace for ~10k examples, increases smoothly with increase

with such an error, the models started to work on the new data

on validation differently, you have to go through the options.

I don't reveal algorithms anymore, I'm just communicating.

I'm just interested in the information beyond training. You can train very well, but it's just learning and fitting...

 
Aleksey Vyazmikin:

What I'm interested in is information beyond training. You can teach extremely well, but it's just rote learning and fitting...

I'm happy with the test results so far

The real tests will show the rest.

I don't want to show the account on the exchange, I don't want to do that because I traded there with my hands... I may show you the different parts later
 
Maxim Dmitrievsky:

I'm satisfied with the results of the tests so far

Real tests will show the rest

My account at the exchange, I do not want to make it public because I traded there mostly with my hands... maybe later I will show you the parts where the bot worked

How often will you re-train the model?

 
Aleksey Vyazmikin:

How often will you retrain the model?

as the situation warrants

When it comes to real trading, you need to keep your eyes open
 
Maxim Dmitrievsky:

by the situation

How do you define the "situation"?

 
Aleksey Vyazmikin:

How do you determine the onset of the "situation"?

By the demolition of the results of transactions in the negative side, perhaps neatness\disabling before the news, retraining after them seems also quite normal, because the market conditions are changing

 
Maxim Dmitrievsky:

on the demolition of the performance of trades in the negative side, perhaps neatness\disabling before the news, retraining after them seems quite normal, as the market conditions change

Will there be a window a day, a week or a fixed drawdown?

What news do you consider as significant? If I look at the calendar, there will be some statistical news, while geopolitical news is almost impossible to follow...

 
Aleksey Vyazmikin:

Will there be a window a day, a week or a fixed drawdown?

What news do you consider significant? According to the calendar, statistical news is constantly somewhere out, and geopolitical news is not really easy to track...

I don't know about the situation.

Mainly news from the states, partly from Europe, like Brexit, UDP, etc. The rest have little effect on the futures... eurofuch I'll limit myself to

1-3 times a month, like nonfarm, what else... trade balance. I don't have a lot of shock news.

The market usually discounts them to the moment of news release, so nothing critical may happen.

 
Maxim Dmitrievsky:

I don't know about the situation.

Mostly news from the states, partly from Europe, like Brexit, UDP, etc. others do not affect the futures very much... eurofuch I will limit myself to

1-3 times a month, like nonfarm, what else... trade balance. I don't have a lot of shock news.

And more often the market discounts them by the time they are released, so there may be nothing criminal going on.

I'm just preparing news predictors for the dollar/ruble. Do you have your Expert Advisor on Moex right now?

 
Aleksey Vyazmikin:

I'm just preparing news predictors for the dollar/ruble. Have you launched an Expert Advisor on Moex now?

Yes, but I haven't done anything with the news, I have to conduct a re-survey, it is an interesting subject. I have not done anything yet, but I have to re-calculate it.