Machine learning in trading: theory, models, practice and algo-trading - page 2140

 

Studied models on data collected on all ticks and OHLC model - the results differ almost twice - I don't understand. Found out that different number of trades, started studying and this is what I found.

Forum on trading, automated trading systems and strategy testing

New MetaTrader 5 build 2690: Improvements in MetaEditor

Aleksey Vyazmikin, 2020.11.19 22:42

What a realistic testing on all ticks on the real data - even the broker's server failure is simulated - stops do not trigger - Otkritie Broker!



And how can you say here that testing on all ticks is more reliable - no way.

 
Uladzimir Izerski:

It is not a question of usefulness or uselessness per se. The question is about the profitability of this or that model.

I see how everyone was excited by my posts on this resource and involuntarily moved the discussion to this thread. But the behavior of some individuals and the nature of communication does not allow me to develop the topic to the end. I am ready to have a dialogue with the adequate ones.

Let's define the terms, what is "model profitability"?

 
mytarmailS:

will do if the question of invariance is solved

And how do you propose to solve it? I divide the speed by my ATR.

 
mytarmailS:

Aaaah, no, that's just not helpful, why do I need zz? to check the strategy with trend lines, and translate everything into symbols, and so what? )))

In addition, I know how to do it, I was just curious to hear more options ...


Well, I do not know... I hastily finished the first draft so to speak algorithm, the generalization is not very good but sometimes something does guess.

If someone is going to poke around in it, we can discuss how to represent the data so that AMO understands something, although I haven't even tried it directly without normalization, maybe it will work that way.

And it seems to work well.

 
Aleksey Vyazmikin:

Let's define the terms, what is "model profitability"?

For example. There are facial recognition programs. How do they work?

They take certain facial points and combine them into one mathematical model.

That's what you should do here in the marketplace. You have to build a rigorous mathematical model, not a blurry blur of indicators.

This can only be achieved by working directly with the price.

 
Uladzimir Izerski:

For example. There are facial recognition programs. How do they work?

They take certain facial points and combine them into one mathematical model.

That's what you should do here in the market. You have to build a rigorous mathematical model, not a blurry blur of indicators.

This can only be achieved by working directly with the price.

What points to take. On the face and on the figure the points are clear. On BP the extremums of points are clear, but there are a lot of them and their formation is random. How to find the points.

 
Valeriy Yastremskiy:

What points to take. On the face, on the figure, the points are clear. On BP the extremums of the points, but there are many of them, and their formation is random. How to find the points.

There is nothing random on the financial charts. Everything is exceptionally consistent from monthly charts to ticks. The wave principle is helpful. You have to work with it.

The chart is cut into waves and manipulated with them. Everything is simple and lies on the surface before our eyes.

 
Aleksey Vyazmikin:

And how do you propose to solve it? I divide the speed by my ATR.

What the hell is ATR?)) I'm talking about preprocessing that allows you to give the model a minute or even a day of data and it will always see the same thing ...

I'll draw and show you how to do it, it's useless to explain.

Aleksey V yazmikin:

It seems to be working fine.

If i've noticed that i have a sloppy time frame, it works more or less well.

 
mytarmailS:

What do you mean by that?

Example - Ishimoku.

 
Valeriy Yastremskiy:


I am planning the logic of the channel breakthrough depending on the state of the series and the previous states.) I plan the logic of breaking through the channel depending on the state of the series and the previous states.

It would be better to have 144.