Machine learning in trading: theory, models, practice and algo-trading - page 625

New comment
 
Elibrarius:

Why do you take it so far away? I thought you had to take the one next door...

I take any, it's just as an example... to be more similar to the original row, for example
 
I take it you are engaged in regression and price forecasting?
My first experiments with classification made me bored... I was thinking about switching to forecasting.
 
Elibrarius:
As far as I know you perform regression and price forecasting?
My first experiments with classification made me homesick... I'm thinking about switching to forecasting.

All at once )) alternately

I have three systems at the same time, now I take one and now the other. Regression is very convenient to build a spread between price and forecast and see the model quality even by eye

 
Max, respect. Cool prediction.... I understand it predicts 1 bar ahead? And a question, is it built on the RF or on a neural network?
 
Anatolii Zainchkovskii:
Max, respect. Cool prediction.... I understand it predicts 1 bar ahead? And the question, is it built on RF or on a neural network?
This one is on RF... how to explain... in fact it shows the spread between 2 pairs, and instead of a given formula it is counted through RF (usually done through linear regression)
 

I'm afraid that regression/prediction on the net will produce about the same thing as looking for similar sites/patterns in history (which I did 3 months ago):

I get the impression that NS has nothing to teach at all, because there are almost no typical situations in the market. Either specially trained robots with big money specifically break patterns.
My pattern finder couldn't determine a typical double top because in the last year in 20 most similar cases the price worked out a double top pattern and flew further up.
The forecast is represented by the bold red (high) and white (low) lines, gray and dark red variants from the history.

Yes, and the most similar variants, not so similar.

Sometimes it predicts in one direction and the price goes in the opposite direction.


So it turns out 50/50, like with a coin.
 
Maxim Dmitrievsky:
This one on RF... how to explain... in fact it shows the spread between 2 pairs, and instead of a given formula it is counted via RF (usually it is done via linear regression)
If i understand correctly, then this video shows how RF adjusts weights for currency pairs instead of a regression approach? the weights change very slowly... that's why it turns out that the forward is long in the zone of truthful calculation...
 
Anatolii Zainchkovskii:
If I understood correctly, then this video shows how RF adjusts weights for currency pairs instead of a regression approach? the weights change very slowly... that's why it turns out that the forward is long in the zone of true calculation...
well yeah, it's just a non-linear model instead of a linear one
 
Maxim Dmitrievsky:
Well, yes, it's just a non-linear model instead of a linear one.

I looked at your solution using RF multiplication table, it turns out that 500 trees is not enough to give the exact answer, and there are no complex calculations... what about the behavior of curves, do I have to model ten thousand trees? And the noisiness of market data makes itself felt...

 
Anatolii Zainchkovskii:

I looked at your solution using RF multiplication table, it turns out that even 500 trees are not enough to give an exact answer, and there are no complex calculations. What about the behavior of curves, is it thousands of 10 trees to model then? And the noisiness of market data makes itself felt...

Why, no, there's an excess of trees... it gave good answers with 10 (I don't remember how many I set)

500 is a lot, enough for any dataset.

but with 10 trees, everything is fine

2018.01.29 21:02:59.333 RF sample (GBPUSD,H1)   Info=1   RMSE Error=0.00415
2018.01.29 21:02:59.335 RF sample (GBPUSD,H1)   Тест 1 >> 9*1=9 // 9*1=9 // 3*1=3 // 5*7=35 // 9*1=9 // 1*9=9 // 3*1=3 // 7*5=35 // 5*2=10 // 1*8=9 // 
2018.01.29 21:02:59.335 RF sample (GBPUSD,H1)   Тест 2 >> 8.3*3.7=32.00(30.71) // 6.4*1.2=6.00(7.68) // 4.0*5.9=24.00(23.60) // 3.1*4.1=11.70(12.71) // 6.4*1.8=12.20(11.52) //
1...618619620621622623624625626627628629630631632...3516
New comment
Reason: