Machine learning in trading: theory, models, practice and algo-trading - page 1609
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
This is called model stacking. It will not be the same, but not the fact that it will be more effective. I did so, did not see any improvement.
There is another way, it's called meta-learning. You train the first model to predict classes, then you get the results and feed them into the second model, on the same or different preictors, which allows/denies trading of the first model. 1 - trade, 0 - do not trade, depending on the quality of the predictions of the first model, i.e. a kind of filter. This greatly reduces the number of errors on the training data, but not so much on the new data (if the model has low generalizability). But in itself meta-training is a normal thing.
You can train the first model on some data, and the meta model on other data, on the errors of the first. There may be different variants. I did it both ways, in general there is an improvement, but it is more of a tweaking, rather than a way to get a quality model that will work on the OOS.
You can google Marcos Lopez De Prado "meta learning", just about trading.
Thanks for the info, but I considered splitting predictors into parts as a way to save computer resources during training, I just can't train 10 predictors in the model at once...
I don't want to make predictors smaller with PCA or something else, because I need to choose important predictors first, so I'm thinking how to break general model into PDP-models with minimal loss of information
Thanks for the information, but I considered splitting predictors into parts as a way to save computer resources during training, I just can't afford to train for example 10 predictors in the model at once...
I don't want to decrease size of predictors with PCA or something else, as I need to select significant predictors first, so I'm thinking how to split general model into POD models with minimal loss of information
What 10 predictors? Remove the correlated ones, see the importance of the remaining ones through forest or boosting and there will be 3-10
You can't guess 100% anyway, it's been tested, there will be incorrect predictions in any case. That's because the models for training can be repeated, but the outcome will not necessarily be the same.
For example what it looks like. Of course I may not have a very good approach to selecting models (formalized data), but I find it hard to believe the possibility of 100% correct predictions.
What 10 predictors? Remove the correlated ones, see the importance of the remaining ones through the forest or boosting and there will be 3-10
What if the predictors are logical rules? :)
What if predictors are logical rules? :)
what does it matter, there are not so many predictors, it's not a model, it's garbage
maybe somewhere in search engines or image analysis, but not for quotes for sure
In what form do you feed the neural network levels?
Who cares, there are not that many predictors, it's not a model, it's garbage
Why?
The "richer" the model, the worse it is?
Especially if you do not know yourself which combination of predictors is better, would it not be correct to feed all possible options into the model and then see the importance of predictors in terms of the modelIn what form do you feed the levels of the neural network?
No no no, I tried to use a two-level, but I did not get the expected result, as I wrote above, maybe not the optimal selection of data (in some cases contradictory) did not allow to see any hint of a logically correctly interpreted result. So far only the usual layered neuronics. Before we start a multilayer network we need to understand if each layer gives the right solution individually.
No no no, tried to use two-level, but I did not get the expected result, as I wrote above, maybe not the optimal selection of data (in some cases contradictory) did not allow to see any hint of a logically correct result. So far only the usual layered neuronics. Before we start a multilevel network we need to understand if each level individually gives the right solution.
sorry i meant support and resistance levels