Machine learning in trading: theory, models, practice and algo-trading - page 1906

 
Maxim Dmitrievsky:

Here's one. There are only 12 fiches, not 24

I have errors

>>> print(train_score, " ", tst_score)

1.0 0.5454545454545454

Redo the file. The target is only two classes, which you can choose yourself, but one 0 the other 1 and the name of the columns nominate from 1. Because I last time got tired of jumping with it. Do it, I'm waiting...
 
Mihail Marchukajtes:
Redo the file. The target has only two classes, which you can choose by yourself, but one is 0 and the other one is 1. I got tired of jumping with it last time. Do it, I'm waiting...

I need three classes.

 
Maxim Dmitrievsky:

I see, you're an optimizer ... I do not understand what you're writing ))

Well look VTRIT left all 24 inputs. If you dug deeper Optimizer Reshetov, then you would understand that he uses a wrap, when the learning process assesses the result of adding or removing inputs. I go from top to bottom, i.e. first I feed 11 inputs, then comes the remove function, which removes one input at a time and evaluates the result, if it improved the input was crap, then comes the add function, this is when he adds one of the unused inputs and if it leads to improvement he leaves it, and the chang function, there are them in the picture above. This function simply removes one and adds one and sees the result. I'm sure you haven't fully reproduced the logic of the Optimizer.

I think I will use python and will repeating it all, just like Reshetov did and with my own tweaks....

 
Maxim Dmitrievsky:

I need 3 classes

Look, this is essentially a Trend Optimizer. Trend up, trend down and flat. The optimizer replies Yes, no, I don't know. You have to choose two classes up and down. And you will have the flat one as an estimator and I don't know how much it will give out. When the flat, + - 10 pips can be and in this case, if the model says I do not know and really candle was without a body, I know it works ....
 

Man, I've just had the VTRIT released... Beautiful :-)

I am very attentive to those signals when the model says I DO NOT KNOW, as a rule this signal leads to the closure of all positions and it is not rare that this signal closed with only symbolic profit, at such moments I am just confident in the model performance and next coming signals usually come as a winner !!!!

 
In short, I just compared the tree with AdaBoost. It's undoubtedly better... but I don't know how to parse it into code yet
 
Maxim Dmitrievsky:
I was just comparing the tree with AdaBoost. It's undoubtedly better... but I don't know how to parse it into code yet
Just send me the last file like I asked... I'll just finish mine here and twist yours.
 
Mihail Marchukajtes:
Please send me the last file as I ask... I'll finish mine here and twist yours.

I need 3 classes

 
Maxim Dmitrievsky:

I need three classes

Maxim, are you stupid? I said that the third class will be a test, and it will be the same during real work. I had an option to improve the optimizer so that the third class could be trained for different values of polynomials. But it can be used as a test class, and it will be used in real work, too... Whatever you want....
 
Mihail Marchukajtes:
Maximka are you stupid? I told you that the third class will be a test and in real work it will be too. I had an option to modify the optimizer so that the third class could be trained for different values of polynomials. But it can be used as a test class, and it will be used in real work, too... Whatever you want....

Ya clever girl