Machine learning in trading: theory, models, practice and algo-trading - page 1843
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
You do not think, no one closes your mouth is just gathered here specific guys who can not tolerate demagoguery. Although AI systems have the ability to produce a non-obvious answer, but it is still an exact science where 1+1=2, rather than approximately, allegedly or not. So also the market is quite a specific type of activity where there are actual news, and there are pseudo teachings. Such as Yusuf's, for example.
Let's turn to Wikipedia, do you trust it?
Market- a set of processes and procedures for the exchange between buyers (consumers) and sellers(suppliers) of certain goods and services.
So maybe information about the relationship between buyers and sellers is important, rather than waves, stochastics, Bollinger, etc.? What do you think? You know, there are a lot of pseudo-experiments that are trying to take the first place.... I think I'll make another video, but this time it's a text. And with the sound, I think we need to come up with something normal. Damn, no one knows why gopro writes sound with noise????
Even read excuses)).
I even read the excuses)).
You know, a thought occurred to me the other day.
The quality of the video I posted leaves a lot to be desired, but judging by the views no one is interested in high matter and scientific reasoning. Everyone is interested to look at the beautiful chick who spins half an hour about the abolition of self-isolation in Gorky Park. She literally in front of my eyes for a day collected 500,000 views against my 70. Conclusion: Scientific topics are not included these days. The main thing is a pretty face and a pleasant voice, and no one is interested in what is there with AI. Too bad :-(
I want to ask everyone's advice.
In the Darch package I found the following option of model evaluation:
We count the error on the track and on the oob sections.
Then the final error is counted as
err = oob_error * comb_err_tr + train_err * (1 - comb_err_tr);
In my opinion, training should be controlled by the average value of completeness and accuracy indicators for the whole sample, dividing the sample into windows of, say, 10%-20%. I myself select leaves by this method, but I also take into account the financial result.
You know, a thought occurred to me the other day.
The quality of the video I posted leaves a lot to be desired, but judging by the views no one is interested in high matter and scientific reasoning. Everyone is interested to look at the beautiful chick who spins half an hour about the abolition of self-isolation in Gorky Park. She literally in front of my eyes for a day collected 500,000 views against my 70. Conclusion: Scientific topics are not included these days. The main thing is a pretty face and a pleasant voice, and no one is interested in what is there with AI. Too bad :-(
Nobody wants the truth, Misha )). Everybody wants beautiful illusions.
You don't have to be a beautiful pacifier yourself, just create an idea in the spectator's mind, turn on his imagination, and then tell the truth.)
Only the sound of a jackhammer in the background bothers me )
Nobody wants the truth, Misha )) Everybody wants beautiful illusions.
You don't have to be a pretty nipple yourself. You have to create an idea in your viewer's mind, turn on his imagination, and then cut the truth.)
Only the sound of a jackhammer in the background bothers me )
In my opinion, training should be controlled by the average value of completeness and accuracy in the entire sample, dividing the sample into windows of, say, 10%-20%. I myself select leaves by this method, but I also take into account the financial result.
This is cross validation. Or you can also roll a forward, so that the validation set is always later than the training set.
Let's clarify the terminology:
- accuracy, you mean standard Accuracy (proportion of correctly classified examples)
- completeness. Is this the number of examples/sample size for training? How do you select it? By selection?
This is cross validation. Or you can also roll a forward, so that the validation set is always later than the training set.
It is important to check for signal robustness over the entire sample, rather than looking at the final score - there can be different methods, one of them I briefly described.
Let's clarify the terminology:
is accuracy, you mean standard Accuracy (the proportion of correctly classified examples)
is completeness. Is this the number of examples/sample size for training? How do you select it? By selection?
Precision is accuracy and Recall is completeness. These indexes are important if there is more than one class and if one class from the set is a signal one. For example, if there is a triple classification - buy(1)/wait(0)/sell(-1) or if you look for volatility - there will be a strong (1) movement or a weak one (0). If in logic the two classes are equivalent, then the meaning of these indicators gets a little blurred.
What about Maximka? Have you read anything? Any cuts?
I improved my approach a little, the results are better too... All inputs showed + :))
But there are problems...
1) there are not enough signals.
2) the model dies in time
But I think I have started to understand something in this damn market, and the breakthrough is not far off))
It is important to check for signal stability throughout the entire sample, rather than looking at the final figure - there may be different methods, one of which I have briefly described.
Precision is accuracy and Recall is completeness. These indicators are important if there is more than one class, and if the signal is one class of the set. For example, if there is a triple classification - buy(1)/wait(0)/sell(-1) or volatility search - there will be a strong (1) movement or a weak one (0). If the logic of the two classes is equivalent, then the meaning of these indicators is a bit blurred.
And in general, Precision can be considered a basic metric when you have a "wait" class. Errors in Precision are direct losses from misclassification.
And Recall means lost profits, i.e., we waited instead of acting.
The bottom line is to maximize F1, which will find the best value with a minimum of prediction errors and a minimum of missed profits.