Machine learning in trading: theory, models, practice and algo-trading - page 1696

New comment
 
Mihail Marchukajtes:

If you're already up to R, why don't you start training models in it? Then you won't need any Excel or other crutches! That's what the fuck it's for!

 
mytarmailS:

If you're already up to R, why don't you start training models in it? Then you won't need any Excel or other crutches!

Well, now we're getting to the fun part. Do you think Reshetov's optimizer just became a good optimizer just by using anchor vector system? No, it didn't. The point is that it implements a set of algorithms surrounded by a neural network. This is the creation of invariant, and the clever division of the training sample, etc. And I once suggested to try to transfer the logic of JPrediction into the same R, but this idea was not supported, so it all remained the same. With my knowledge, if I spent all day yesterday creating a composite matrix of a certain size, in which I needed to put values from another table based on data from the third one. I only did data.frame for six hours and only at the end saw that it is formed on columns in contrast to the list, which is formed on the rows. I had no one to tell me what to do. And where, by the way, is Doc? I haven't seen him for a long time.....
 
And I'm not even talking about my added extras, which significantly reduce optimization time and increase the likelihood of getting exactly the generalized model.
 
Mihail Marchukajtes:
Well, now we got to the most interesting part. Do you think Reshetov's optimizer just became a good optimizer by stupidly using a system of reference vectors? No, it didn't. The point is that it implements a set of algorithms surrounded by a neural network. This is the creation of invariant, and the clever division of the training sample, etc. And I once suggested to try to transfer the logic of JPrediction into the same R, but this idea was not supported, so it all remained the same. With my knowledge, if I spent all day yesterday creating a composite matrix of a certain size, in which I needed to put values from another table based on the data from the third one. I only did data.frame for six hours and only at the end saw that it is formed on columns in contrast to the list, which is formed on the rows. I had no one to tell me what to do. And where, by the way, is Doc? I haven't seen him for a long time.....

Shit)))....

How do you know it's good? Did you compare it to anything else? What did you compare it to?

 
mytarmailS:

Shit)))....

What makes you think it's good? Did you compare it to anything else? What did you compare it to?

Unfortunately, if you remember, all my attempts to make a comparative analysis with other models, including models from R, failed. Because my dataset was too small for the masters of toutos. However, to make a comparison experiment with sufficient purity it is necessary to copy the algorithm of JPrediction in R and conduct a comparative analysis. And then you can already finetune training methods and types of networks to change here and there. But so far I judge the optimizer's work only empirically. Stupidly through practice. If it (the optimizer) in fact is a fudge, then I can not even imagine what stunning results can be obtained in other systems, if this fudge quite helps me to earn.
 
Mihail Marchukajtes:
Unfortunately, if you remember, all my attempts to make a comparative analysis with other models, including R models, failed. Because my dataset was too small for the gentlemen here. However, to make a comparison experiment with sufficient purity it is necessary to copy the algorithm of JPrediction in R and conduct a comparative analysis. And then you can already finetune training methods and types of networks to change here and there. But so far I judge the optimizer's work only empirically. Stupidly through practice. If it (the optimizer) in fact is a dummy, then I can not even imagine what stunning results can be obtained in other systems, if this dummy quite helps me to earn.

what exactly is the problem with comparing JPrediction to something else?

 
mytarmailS:

what exactly is the problem with comparing JPrediction to something else?

In general, the problem of comparison was too small training set. After all, the method of reference vectors is labor-intensive because it shakes the data thoroughly. And the idea was the following. We take the same training file and get two models - one in the optimizer and the other in the alternative system - and then just compare their performance on the feedback loop. If you're interested, we can try it...
 

So there you go. It didn't take long for the epicfeel to arrive. Once again I am convinced that money loves silence. Although right now I see a good opportunity to stand at a better price. Shorting a little bit, probably, not for a long time, within the limits of today.


 
mytarmailS:

what exactly is the problem with comparing JPrediction to something else?

If you want, we can fiddle around in what way. I'll write a series of key conditions implemented in the optimizer to create the system. You make it in R, but with free interpreting. Then we both train the same file and see how it works on the OOS. Maybe you already have your own AI in R, different from Reshetov's optimizer logic, but you'll need to define the rules of the game, so to speak, so that the comparison was adequate, rather than compare elephant with a fly. There and we will dot all the I's and cross all the t's!!!!
 
The most interesting thing is that this growth in SI begins to win back the next option, well, what can I say. Well done...
1...168916901691169216931694169516961697169816991700170117021703...3494
New comment
Reason: