Machine learning in trading: theory, models, practice and algo-trading - page 2985
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
I wonder, is there any sense in trying to shove these memory-based into ONNX file, or is it better to do the calculation by MQL means?
If it will be in the model at once, it will be easier to change them.
By the way, is there a possibility to use more than one model in an Expert Advisor, let's say, to select them without changing the code?
I haven't analysed this embedding yet.
In the sense in which KNN and LWLR are local . The output depends only on close points, not on the whole trayne.
What if the trayne is reduced to the required size?
Well, then you can add individual decision trees to your candidate list.
Still, each tree is global, since it counts on the whole traine.
What if we shrink the trayne to the right size?
Yeah, that's the basic idea for now.
Still, each tree is global, as it counts on the entire traina.
I wasn't paying attention, I missed the treyne.
If it's right away in the model, it's easier to change them.
If we are talking about direct VPS trading, then probably yes. But if, for example, you need to add a line to the table stored in the ONNX file for KNN over time, then there is no way. Although, perhaps, it is possible to use this table as an input for ONNX model. It is necessary to understand whether it makes sense to use ONNX here at all.
By the way, is it possible to use more than one model in an Expert Advisor, say, to select them without changing the code?
There seems to be no problem with it - several models are opened, each with its own handle. Although, there will probably be problems with multi-gigabyte models).
If we are talking about direct VPS trading, then probably yes. But if, for example, you need to add a line to the table stored in the ONNX file for KNN over time, then there is no way. Although, perhaps, it is possible to use this table as an input for ONNX model. It is necessary to understand whether it makes sense to use ONNX here at all.
There seems to be no problem with it - several models are opened, each with its own handle. Although, with multi-gigabyte models there will probably be problems)
Is the data stored there in its original form so you can just add it? I thought that a normal model has data transformation and sorting with indexing to speed up work. If there is, then the model just needs to be rebuilt every week or month, or maybe even day - with periodicity.
If there is nothing like that, and the whole database is loaded into memory to speed up the work, then yes - you can make automatic addition of examples and here it is better to do everything in MQL. There is an article here on how to use such models in MQL.
Is the data stored there in its original form so that it can be added? I thought that a normal model has data transformation and sorting with indexing to speed up work. If there is, then the model just needs to be rebuilt every week or month, or maybe even day - with periodicity.
If there is nothing like that, and the whole database is loaded into memory to speed up the work, then yes - you can make automatic addition of examples and here it is better to do everything in MQL. There is an article here on how to use such models in MQL.
How to do it using MQL5 is more or less clear. How to do it via ONNX is not very clear yet.
By the way, I found an attempt to run KNN model in ONNX format on the forum. But, as I understand, it has not worked yet.
It is more or less clear how to do it using MQL5. How to do it via ONNX is not very clear yet.
By the way, I found an attempt to run KNN model in ONNX format on the forum. But, as I understand, it has not worked yet.
Well, the sooner the initiative is shown, the sooner the search for error correction will begin.
And what do you find the advantage of such models? Have they proven themselves better than others? More robust in theory to changes in input data?
If it's right away in the model, it's easier to change them.
By the way, is there any possibility to use more than one model in the Expert Advisor, let's say, to select them without changing the code?
Why not? We should publish such an example soon.
There you go - the sooner initiative is taken, the sooner the search for bug fixes will begin.
What do you find the advantage of such models? Have they performed better than others? Are they more resistant in theory to changes in input data?
Simplicity of implementation and ease of retraining/self-learning, when you can just add new rows and delete old ones.
There are plenty of drawbacks too - it doesn't work well with a large number of attributes, for example. But as a way of initial evaluation of the usefulness of features it's fine.
The interest in having some local analogue of trees comes from the fact that they are more suitable for discontinuous patterns, while KNN and LWLR are more suitable for continuous ones. One would like a more complete set of tools, so to speak.