Machine learning in trading: theory, models, practice and algo-trading - page 2771

 

The result of receiving signals on EURUSD from 2021.03.01,00:00.

There is no training sample and no "out of training" sample.

The model works on H1.

Datamining and training of the model at each hour.

The initial number of predictors is about 170.

Number of MO= 3 models using 5 to 10 datamined predictors.

Run 1000 bars. Calculation time = 6.78 hours.

Predicts short, long and out of market - ternary model

Prediction results:

Prediction 1 hour ahead

short_er <- 0.8947

long_er <- 0.9285


Prediction for 2 hours ahead

short_er <- 0,6578

long_er <- 0,8928

Prediction 3 hours ahead

short_er < -0, 5789

long_er <- 0, 6785


 

Exact prediction for 5 (!!!) hours ahead: the rate will not change...there is no trading.

 
Maxim Dmitrievsky #:

Curious note in favour of Sanych's retraining and stuttering window

https://medium.com/towards-data-science/ai-in-industry-why-you-should-synchronize-features-in-time-series-f8a2e831f87

It's a known fact: it's best not to trade the first few hours on Monday. By the way, so are the last few hours on Friday.

By the way, your EA


 
СанСаныч Фоменко #:

Result of signals on EURUSD since 2021.03.01,00:00:00

No training sample and no "out of training" sample.

The model works on H1.

Datamining and training of the model at every hour.

The initial number of predictors is about 170.

Number of MO= 3 models using 5 to 10 datamined predictors.

Run 1000 bars. Calculation time = 6.78 hours.

Predicts short, long and out of market - ternary model

Prediction results:

1 hour ahead prediction

short_er <- 0.8947

long_er <- 0.9285


Prediction 2 hours ahead

short_er <- 0,6578

long_er <- 0,8928

Prediction 3 hours ahead

short_er < -0, 5789

long_er < -0, 6785


Can I see the trades?
 
mytarmailS #:
Can we see the deals?

There's no ready-made, you have to move your hands. Much more important is a radical reduction in time. 1000 bar is nothing. That's the main problem today.


I put it out there for the sake of campaigning on the predictive power of predictors.

 
mytarmailS #:
Can we see the deals?

Thought I'd look into it, it's not easy for me, but really very interesting.

 
СанСаныч Фоменко #:

There is no ready-made, you have to move your hands. Much more important is a radical reduction in time. 1000 bar is nothing. That's the main problem today.


I put it out there to agitate for the predictive power of predictors.

What's not available?
 
mytarmailS #:
What is not ready?

Binding to the quote chart.

 
СанСаныч Фоменко #:

Bindings to the quote chart.

Was there any trading going on? Or was it all at the level of tests in the r-ke?

 
mytarmailS #:

Was there any trading going on? Or was it all at the level of tests in the r-ka?

Just R.