Machine learning in trading: theory, models, practice and algo-trading - page 2590
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
Earlier you had the idea of combining standard metrics with custom metrics, which I understood as training models by standard metrics and selecting by custom metrics.
yes, in fact, the combined criteria is better
training by standard metrics, but there are so called Evaluation metrics like asscgasu or roc, bussing is stopped by them, though it is trained through logloss minimization. Then we select by our custom metrics.
Then you can use models in the MT5 optimizer for tweaking parameters of stop loss, etc. And there is a lot of thankless work :)
In the parameter space of the model? It has an enormous dimensionality. This is possible only for very simple models with a small number of predictors.
It is not very clear how you can build a surface in the space of huge dimensionality. We simply have very few points in comparison with this dimensionality. Unless by some way of visualization with dimensionality reduction like PCA and so on, but the meaning is not clear.
Then you can run the models in the MT5 optimizer to adjust the parameters of stop loss, etc.
What should I do it with? Or are we talking about the models written in MQL5?
What do you use to do that? Or are we talking about models written in MQL5?
Or transferred, see articles about catboost on this site (search). Mine or Alexey's.
Mm, I haven't got the gist of how python was ported to MQL5, but I realized that apparently using the function fromhttps://www.mql5.com/ru/articles/8642 you can use catbust models from mql5 code. I mostly use XGBoost, but catbust is also possible.) I agree about neural networks and tabular data. As far as I understand, using code from article I can use Catbust models, but other libraries can't. That's good too. Now I use approach where I'm not limited in my choice of Python libraries - I'm pulling models from mql5 code through Python API. In this case I'm not limited by libraries, but it's slower; the variant from the article has limitation in libraries, but apparently it's much faster.
Yes, that's exactly how I see it... Reconstruct the OP and then visualize it through PCA...
To be honest, it doesn't make much sense. Small-dimensional projection in principle can not display everything that can be in the multidimensional case. Extrema, for example, can be saddle points - a maximum in some variables and a minimum in others. And if the surface is not smooth, anything can be there.
Theoretically, it might make a bit more sense to use PLS instead of PCA.To be honest, it doesn't make much sense. A small-dimensional projection cannot, in principle, represent everything that can be in the multidimensional case. Extrema, for example, can be saddle points - a maximum in some variables and a minimum in others. And if the surface is not smooth, anything can be there at all.
Theoretically, it might make a bit more sense if you use PLS instead of PCA.Maybe the current implementation makes little sense, but it may make more sense in the essence of what it's all about than in this entire branch
Mm, I haven't got the gist of how python was ported to MQL5, but I figured out that apparently using the function fromhttps://www.mql5.com/ru/articles/8642 you can pull catbust models from mql5 code. I mostly use XGBoost, but catbust is also possible.) I agree about neural networks and tabular data. As far as I understand, using code from article I can use Catbust models, but other libraries can't. That's good too. Now I use approach where I'm not limited in my choice of Python libraries - I'm pulling models from mql5 code through Python API. In this case I'm not limited by libraries, but it's slower, in the variant from the article - limited by libraries, but apparently order/order is faster.
The variant from the article is already final, for applying python-trained models in MT5 without layers. It saves tree splits in c++ from python program and then translates them into mql code, plus class calculation function.
Great, I'll have to try it sometime.
Great, I'll have to try it sometime.