Machine learning in trading: theory, models, practice and algo-trading - page 1221

 
Vizard_:

)))

Why?

 
The main thing is not to launch them in the real world:

I would be glad to be wrong, but IMHO "another approach" does not exist, in this case everything is mainly determined by data, their quantity and quality, with "pure Forex" even 5% correlation of the forecast with returnee is difficult to squeeze out, it certainly may be enough, but annualized SR will be ~2, it is exciting, anxious, you can lose a healthy sleep. But there are 100500 ways to retune or just staring at some defective models showing cosmos, it's good for the nervous system, but the main thing is not to run them for real)))

So with the reliable data "on the purchase" can be without MO, and why all about the data and not a word about quality models that can out of this pile of crap... Why are there no such models?
 

Well, we don't have any data, I agree... except for broken DT quotes.

We do heroic things to get the working data out of them.

 
toxic:

the MO models are the most common, forests, gradient boosting, with the right features and targeting on hundreds of thousands of points, the fit is minimal

since the models are the most ordinary, it is obvious that you are among those who do not give them any importance - what they gave, that's what we drive....
 
Vizard_:

(Just wanted to know, asked, found out, and that's all...)) you should have said, now he will zapet levels, or even worse - will make you look at their))) hilarious...

Well you can see that I don't even develop the topic and don't ask for more details... I just wanted to know, asked, found out... that's all)

 
toxic:

I used to attach great importance to it, I rewrote most algorithms with my own hand, wrote dozens of my own versions, but as it turned out it was only to raise my skills in MO, as a result any nub can (using the same data and chips) configure XGB with some gene-optimizer and get similar results. The advantage is not in what classifier to take and how to configure it, you need more data and quality features, and infrastructure to make it all work comfortably without confusion and agony. And then you need to know how to trade it all.

The fact is that the search for ways to reliable data, the topic is very old and is unlikely to be any progress in it, rather the opposite, which can not be said about the development of MO models.
Why to fixate on busting, where models are tailored for BP the same recurrent networks with LSTM, there are already generative networks that create data for training themselves, and we all look for the insider:)
 
Ivan Negreshniy:
The matter is that the search of ways to reliable data, the topic is very old and is unlikely to be any progress in it, rather the opposite, which cannot be said about the development of MI models.
Why to dwell on boosting, where models are tailored for BP the same recurrent networks with LSTM, there are already generative networks that create the data for training, and we all look for the inside :)

I'm sorry to interrupt your conversation, but...

why is the next line (incomplete!) of your message not carried to the next line? (I have MT4, maybe that's the reason?)

 
aleger:

I'm sorry to interrupt your conversation, but...

why is the next line (incomplete!) of your message not carried to the next line? (I have MT4, maybe that's the reason?)

I don't know, maybe because of MT4, but anyway everything depends on the program, and they give me data, data...:)
 
Ivan Negreshniy:
I don't know, maybe it's because of MT4, but it all depends on the program, and they give me data, data...:)

If it's no secret, what browser do you use? (I have Opera)

 
aleger:

If it's no secret, what browser do you use? (I use Opera)

MS IE 11