Machine learning in trading: theory, models, practice and algo-trading - page 773
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
Opening trade - screenshot, closing - screenshot. In a week, provided,
Both will be there and will not swear - you will get gifts.
So you don't trust the signal service completely.....?????
Strange!!!!
And I have no time and no desire to comment on every trade. The robot chops, the result is on the face!!!!
I have a news feed coming fromR-bloggers. Today came an ad for DALEX, a variable importance selection package. Tried to install - it does not install for my R 3.4.2.
I really liked the idea though.
Usually variable importance is importance in the sense of how often the predictor was used in fitting the model.
But DALEX uses a different idea: predictor importance refers to the effect of that predictor on the success of the prediction . The model itself is treated as a black box.
I've tried to remember all the packages I've used and I can't recall a package with the same idea of influencing the prediction.
Can anyone help me?
I have a news feed coming fromR-bloggers. Today came an ad for DALEX, a variable importance selection package. Tried to install - it does not install for my R 3.4.2.
I really liked the idea though.
Usually variable importance is importance in the sense of how often the predictor was used in fitting the model.
But DALEX uses a different idea: predictor importance refers to the effect of that predictor on the success of the prediction . The model itself is treated as a black box.
I've tried to remember all the packages I've used and I can't recall a package with the same idea of influencing the prediction.
Can someone give me a hint?
Case in point. Hundreds of articles on the Internet about Arima, and everywhere - "find autocorrelation and autoregression," then a dozen pictures, and immediately the answer with three parameters without any explanation. In 10% of articles can even mention that there is seasonality.
Sorry, but this is from technical analysis - you took an indicator, looked it up, liked it, and made an Expert Advisor.
When we try to use statistical models, the initial question before trying to use it is whether the model is applicable to our data.
If we talk about ARIMA, it is a very limited model in its applicability, especially in financial markets. The people who created it understood this limitation and therefore provided it with additional tools that allows you to determine the usability of the model in a specific case. In practice, we have to check the applicability in a WINDOW, so when the window moves, the model can be applied, then it cannot.
But the situation is even worse.
It is not only in the initial data, to which the model, for example ARIMA, may not be applicable. It is also a result of fitting: all the parameters were adjusted, all the parameters were defined, and then we start to get into it and see that the parameters are NOT significant - they are absent, although we can see them.
Here wrote above that "the situation is even worse". And if you compare it with TA, it is a blind man's situation with a sighted man. If to take into account that indicators are autoregression and ARIMA is autoregression too, but one can find out the applicability of ARIMA, while indicators are always used blindly and then we are surprised that the deposit has moved to the blind.
I have a news feed coming fromR-bloggers. Today came an ad for DALEX, a variable importance selection package. Tried to install - it does not install for my R 3.4.2.
I really liked the idea though.
Usually variable importance is importance in the sense of how often the predictor was used in fitting the model.
But DALEX uses a different idea: predictor importance refers to the effect of that predictor on the success of the prediction . The model itself is treated as a black box.
I've tried to remember all the packages I've used and I can't recall a package with the same idea of influencing the prediction.
Can someone give me a hint?
On 3.4.3 installed. Interesting stuff, but the authorship is suspicious.
Have you forgotten about LIME? Remind me about varbvs as well. I'm leaning more and more towards Bayesian methods in everything
Good luck
I.e. usually backtesting, but this time it is forward thinking?
Well, yes! Why do we need a backtest? Why do we need a backtest? For training, I understand, but for backtesting...
Above in this thread, I posted the results for training and forward - just depressing.
But that's only part of the results I have.
I ran all 6 models on 14 currency pairs at 15 000 H1 bars in rattle: half of them were used for training and the other half were used for forward training.
The results are rather disappointing: out of 84 (in reality there are 168 entries (long+short) there are less than ten, and there are no currency pairs with both long and short positions!
Installed on 3.4.3. Interesting stuff, but the authorship is suspicious.
Have you forgotten about LIME? Remind me about varbvs as well. I'm leaning more and more towards Bayesian methods throughout
Good luck
Is there a 3.4.3 in microsoft?
Thank you for LIME.
SanSanych, you are quite a competent and theory-supported person. Tell me, have you tried to feed the coefficients of polynomials from different series as predictors?
No