Machine learning in trading: theory, models, practice and algo-trading - page 947

New comment
 
information only on the arr_Buy column

I see, thank you. it's just that it didn't coincide with arrSell everywhere by logic, that confused me
 
Dr. Trader:
information only on the arr_Buy column

I see, thank you. it's just that with arrSell did not coincide with the logic, it confused me

I was just as tense, but then I realized that there is a filter on arr_Sell. I just copied only the target ones into the old file.

 
Aleksey Vyazmikin:

I was just as tense, but then I realized that there is a filter on arr_Sell. I just copied only the targets in the old file.

Error matrix for the Decision Tree model on Pred_027_2016_H2_T.csv [validate] (counts):

      Predicted
Actual   -1     0 1 Error
    -1 7848   637 0   7.5
    0    85 14055 0   0.6
    1     0  7789 0 100.0

Error matrix for the Decision Tree model on Pred_027_2016_H2_T.csv [validate] (proportions):

      Predicted
Actual   -1    0 1 Error
    -1 25.8  2.1 0   7.5
    0   0.3 46.2 0   0.6
    1   0.0 25.6 0 100.0


Error matrix for the Decision Tree model on Pred_027_2016_H2_T.csv [test] (counts):

      Predicted
Actual   -1     0 1 Error
    -1 7858   613 0   7.2
    0    61 13986 0   0.4
    1     0  7896 0 100.0

Error matrix for the Decision Tree mode l on Pred_027_2016_H2_T.csv [test] (proportions):

      Predicted
Actual   -1  0 1 Error
    -1 25.8  2 0   7.2
    0   0.2 46 0   0.4
    1   0.0 26 0 100.0
 
SanSanych Fomenko:


In words, please.

I do not understand why you do not have a guessing unit?

 
Aleksey Vyazmikin:

In words, please.

I don't understand why you don't have a guessing unit at all?

It's not me - it's rpart.

It's very strange: only -1 is good.

"0" isn't very good either: 13986/(13986+7896+613) =62% guessed or 37.82% error

 
Generally speaking, a 3-class target is not a good idea. It is better to break it into two targets: (-1,0) and (0.1), and then combine them when deciding on a position.
 
SanSanych Fomenko:

It's not me - it's rpart.

Everything is very strange: only -1 is good.

"0" is not good either: 13986/(13986+7896+613) =62% guessed or 37.82% error

Pretty weird, but if you take the 2015 file, will it guess 1?

 
SanSanych Fomenko:
Generally speaking, a target of 3 classes is not a good thing. Better to split into two targets: (-1,0) and (0.1), and then combine them when deciding on a position.

So I was asked to bring it together.

 
Aleksey Vyazmikin:

I've been asked to do the opposite.

Maybe it's a matter of taste...

By the way, the answer to your question:

Variables actually used in tree construction:
[1] arr_Sell

only one variable was used for some reason.

I didn't wait for the other models. I will put a random forest. If I wait, I will post it.

 
SanSanych Fomenko:

Maybe it's a matter of taste...

By the way, the answer to your question:

Only one variable was used for some reason.

Didn't wait for the other models. I will put a random forest. If I do, I'll post it.

So I do not understand, he used a different array, or offered to use another array with the target?

Just the result is a good show, I need to understand what and how to reproduce it :)

1...940941942943944945946947948949950951952953954...3492
New comment
Reason: