Machine learning in trading: theory, models, practice and algo-trading - page 927
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Programmatically the tree can be described like this :
And there are no exact names for the predictors - the picture modes the names just....
Here are the full names:
I will try to do the same for mnogoVhodov_02. I will leave the script to work overnight, and tomorrow afternoon I will show what has come out.
However, the accuracy of the model for malovhodov is not very good, there are a lot of false entries. I would not trade )
I ran my forest for true and false marks.
Looking for a class more than half in another class in the test hit, but in training well divided)
Good result. I do not use a forest, but one tree, which is probably why I have noticeably worse.
And what are the results on the test with the data from another file? (different year)
Here are the full names:
I will try to do the same for mnogoVhodov_02. I will leave the script to work overnight, and tomorrow afternoon I will show what has come out.
However, the model accuracy for malovhodov is not too good, there are a lot of false entries. I would not trade )
Thanks for the full screen of the tree, I will now try this set on a tree from the program.
On the other hand if the tree has something i may use, i have high hopes for the woods, i will try it later, the results should be about 15% according to my expectations and that should be good.
And about few inputs, that's what the filter set is for - the idea is to improve the situation in the aggregate.And what are the results on a test with data from a different file? (different year)
This is my data(one file). The test there is 25%.
Here are the full names:
I will try to do the same for mnogoVhodov_02. I will leave the script to work overnight, and tomorrow afternoon I will show what has come out.
Anyway, the model accuracy for malovhodov is not too good, there are a lot of false entries.)
I have builded a tree in Deductor program and did not find any target.
I see that passions do not subside... In the meantime, a week has passed comparing the two networks of battle. I thought, why am I going to feed you all sorts of tests, etc. The best way to solve the issue is by combat and the result is this...
ELMNN- networks built in R worked for a week like this...
jPrediction- Reshettes like this....
It's hard to judge who is cooler. I think both optimizers are good. But it looks best here.....
And don't rub it in Grandma's face!!!!!!!!
The main thing is that Akello won't miss again next week.
It occurred to me that my basic strategies are crap. Can someone give me some basic strategies that I'll try to improve with my agents?
I tried malovhodov for starters.
Tried to teach forest to predict arr_Buy from 2015 based on arr_Vektor_Week, arr_Vektor_Day, etc.
The classes are very unbalanced (there are 10 times more examples with class 0 than class 1), this adds a lot of difficulty.
This is the 2015 tree trained on
The prediction in both cases is not very accurate, but the accuracy is at least more than 50% in both cases.
I stopped counting the standard errors on these tables.
My reasoning is as follows: the initial class "0" gave a prediction of class "1" = 86118, and the class "1" gave a prediction of class "1" = 12256. This means that when trading, we will get false class predictions = 86118, while correct predictions = 12256, i.e. error = 86116/(86116+12256) = 87.5%9(!!), if class "1" = entry/position, then it is a disaster. And the "0" class position is very good - the erroneous zeros in decision making will be only 5.3%.