Machine learning in trading: theory, models, practice and algo-trading - page 80

New comment
 
What does "RF 10/5" mean?
 
mytarmailS:
We are all demagogues here, only you are d'artagnan, it is clear already, at least you are not a troll.... bye..... :)


I think that the main thing is that the market is stable and the market is stable... I think that the main thing is that the market is stable and the market is stable and the market is stable... If you do not know what to expect from the market, you will not get an answer from me.

This man is Ph. D. in Technical Sciences, has long ago (about 20 years ago) defended his dissertation on "AI". He's been building robots for over 20 years and has a lot of experience.

He states that the market cannot be predicted from a black box position, it is necessary to identify the working attributes, to understand how and why they work and to filter the data to leave only what works, ignoring the noise.

He has about 100 signs (predictors) in his network, each sign has a whole library or package as you want.

And now compare what gulf in quality there is between a sign that requires a whole library and some silly little thing called "SMA". SMA", "MACD" RSI and other ... I don't think there is 0.00000001% useful information in it, likeMihail Marchukajtes wrote, and this is a fact, otherwise models would show exactly the efficiency they can show, which means 90% of correct answers

This man recommends reading "MSUA" and spectral analysis in particular Fourier

========================================================

Further what results I "demagog" have achieved, but in fact very modest, good ideas as I think a lot, my research is going in many directions simultaneously and there is a huge lack of knowledge in various fields, so often ask for help from forum participants, but especially no one wants to help, they say you learn yourself, and then..... only if I myself mastered all then what I this communication, as without logic to me ... well that I digressed


Here's the best thing on the moment. that I managed to squeeze out of RF on the new data is 50% per month for two months in a row, but everything is still very unstable, I tried to upload pictures 10 times, but it does not work (got it)


The bottom line is that there is no need to limit yourself to patterns such as 30% per year is cool, it's not cool, this framework for the mind and creativity

Congratulations. That's how you trade for five years. Or show any data real. statistics for 5 years.

And about the signs, I do not want to make a couch analysis, like, your signs are rubbish, there are better signs, the models are not important. Well, show me what you did. I posted all my inputs here, and their combinations, too. Count their informative value. Why else would you be swaying the air?

 
SanSanych Fomenko:
What does "RF 10/5" mean

Yes, I made a note for myself on the picture when I tested the model on new data, don't pay attention to it ....

this is the model parameter 5 trees 10 branches in a tree

 
mytarmailS:

Yes, I made a note for myself on the picture when I tested the model on new data, don't pay attention to it ....

this is the model parameter 5 trees 10 branches in a tree

Very interesting!

Can you throw in any details?

 
mytarmailS:

Here's the best at the moment. that I managed to squeeze out of RF on the new data is 50% a month for two months in a row, but everything is still very unstable, I tried to upload pictures 10 times, but did not get (got it)


The point is that you do not have to limit yourself to patterns like 30% a year is cool, it's not cool, it's a framework for the mind and creativity

this is a backtest on out-of-sample data? Did I get that right?
 
SanSanych Fomenko:

It's all very interesting!

Can you throw in any details?

What exactly do you want to know? But no one cares. Everyone has his own understanding of the market - his own way. Accordingly, everyone perceives information differently and is not always ready to listen to the other - to listen and not to hear.
 
Alexey Burnakov:
Is this a back-test on out-of-sample data? Did I get it right?
Yes
 
mytarmailS:
yes
develop the idea. add more data. need a couple of years to pass out of the sample. And so, good, well done!
 
mytarmailS:

Sounds like improper model training, too much variation. Models usually have some random processes in training, where the learning logic is not consistently defined. Such random moments lead to the fact that having trained several models, although they will give approximately the same results on training data, but on the fronttest there will be differences.

There are several sources of the problem and its solutions:
1) there are noise inputs which don't carry useful information, they must be removed
2) change training parameters of the model. For neuronkey I solved this problem by using decay parameter (learning curve deceleration), the results on fronttest with this parameter became less scattered. What to do with the forest, I do not know.
3) Make a committee of models. Train a lot of models, fronttest on all models, take the result where most say
4) if you do crossvalidation in the process of training - then repeat it several times on the same data, see how much variation in the results, choose models and predictors with a small spread

This is what came to mind right now, but it's not the limit of possible problems.

 
Alexey Burnakov:
Develop the idea. add more data. need a couple of years out of the sample. But anyway, good job!
there is a ser :)
1...737475767778798081828384858687...3492
New comment
Reason: