Machine learning in trading: theory, models, practice and algo-trading - page 2207

 
Maxim Dmitrievsky:

It says that making adequate markup is expensive and generally unknown... so semi-supervised learning may work better in many cases

it was tested on SEALs and many others, it worked fine

But it's not expensive for us (in our case) to do adequate markup, we can't do it through class labels, but we can describe it through searching for the minimum...

And it should work better ... do you know what I mean?

 
mytarmailS:

But it is not expensive for us (in our case) to make adequate markup, we can't do it through class labels, but we can describe it through searching for the minimum...

And it should work better ... do you know what I mean?

through finding the minimum of what? let's say you minimized the drawdown and got a good profit factor or something, but that doesn't mean that your f-from the minimum, in terms of adoring ability

moreover, you don't know at all how your data is distributed and what's in the new data
 
Maxim Dmitrievsky:

through finding the minimum of what? let's say you minimized the drawdown and got a good profit factor or something, but that does not mean that your f-from the f-from the minimum, in terms of adoring ability

Well do as you did classically, as Ivakhnenka did 50 years ago ))

Three samples, one to synthesize the model, the other two to check.

Maxim Dmitrievsky:
Moreover, you don't know how your data are distributed and what's in the new data

we don't need to know it because we know the labels (we described what's the minimum).

 
mytarmailS:

Well, to do it by the classics, as you and Ivakhnenka did 50 years ago)

Three samples, one to synthesize the model, the other two to check

We don't really need to know that, because we know the labels (we described what the minimum is).

My labels change too, they're random. If I knew the pattern I would not suffer)

 
Only there are no packages in which you can implement your fitness features (((
 
Maxim Dmitrievsky:

My labels change too, they are random. If I knew the pattern, I wouldn't suffer.)

What do you mean?

You do not make a markup on the tray? zigzag or something else, you just randomly poke labels?

 
mytarmailS:

What do you mean?

You don't do the markup on the tray yourself?

I do it randomly. It's all there in the article. Pseudo-random, or rather. Labels are always good (in terms of profitability), you can feed noise into them and make them worse

 
Maxim Dmitrievsky:

I do it randomly. It's all there in the article. Pseudo-random, or rather. Labels are always good (in terms of profitability), they can make noise and make it worse

Labels there (in the article) seem to be meaningfully done, even by people. And you have a random sample of good ones, so pseudo?

adequate markup there is full implied?

 
Valeriy Yastremskiy:

The labels there (in the article) seem to be made meaningfully, even by people. And you have a random sample of good ones, so pseudo?

adequate markup there is full implied?

What article? I mean mine.

I don't really understand the question.

I somehow understood from the beginning that the markup based on zigzags and the like is meaningless.

I can take any profitable signal from local signals, add some attributes and train it. The main thing is to train them to earn, not to sink.

 
Maxim Dmitrievsky:

What article? I mean mine.

I don't really understand the question.

For some reason, I realized from the very beginning that markup by zigzags or other is actually meaningless.

I can take any profitable signal from local signals, add some features and train it. The main thing is to train it to earn, and not to sink.

That link I gave on the wiki about the half... training. I understand the marks are the edges of the stable areas.

If I look at the marks, I see that they are too many examples with different signs and the result cannot be good.