Machine learning in trading: theory, models, practice and algo-trading - page 2967
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've basically started reversing the file format. Everything I've come across before says it's a json compressed protobuf. I just found this: "The ONNX-formatted model is a file in the Protocol Buffers format, which is a message file format developed by Google"
ONNX is first and foremost a language. Here, for example, is a list of its operators. How exactly programme files are represented in it is an important but secondary question. It seems to be a question of having a converter from one language to another.
ONNX is first and foremost a language. Here, for example, is a list of its operators. How exactly programme files are represented in it is an important but secondary issue. Apparently, the issue is the availability of a converter from one language to another.
You can do everything in r, only train the final model in Python, if you need to cram it into the terminal.
you can do it without even installing anything, via google colab.
I used xbox to train it that way :)
On r do everything, the model in python only to train the final one, if it is necessary to cram it into the terminal
you can do it without even installing anything, via google colab
Here it is a matter of violation of the principle that R has everything, almost like Greece) For me it is the first case when there is not something necessary there. Perhaps the language is starting to give up its position.
I used xbox to teach like this :)
Probably it is possible to teach through a network of a TV set with a fridge and an iron).
It's about violating the principle that R has everything, almost like Greece)
Here it's a matter of violating the principle that R has everything, almost like Greece) For me it's the first time when there is not something necessary there. Perhaps the language is starting to lose ground.
Perhaps it is possible to teach through the network of the TV set with a fridge and an iron)
Is onnx really that necessary in R?
Basically ONNX is a sequence of elementary mathematical operations written when executing a model/function/module. Any supported sequence can be translated into ONNX. But since this is all very loosely standardised, it requires extra effort to learn, apply this topic or only apply packages that have inbuilt converters.
I tried in R to convert H2O model saved in MOJO with Python converter (only there are converters from JSOM, MOJO to ONNX) it didn't work. Need to delve deeper into this topic.
I sent a request for the possibility of converting libraries to R (Torch, H2O, XGBoost). Only the last one responded and promised a JSON converter. So far there is no result.
In general, it is necessary to evaluate whether those advantages of ONNX application are worth the efforts to study and use it. This is for everyone to decide.
Of all the converter packages that I have reviewed, the most advanced, well documented and understandable (in my opinion) spox(v0.6.1).
Good luck
Here it is a matter of violating the principle that R has everything, almost like Greece) For me it is the first time when there is not something necessary there. Perhaps the language is starting to lose ground.
Perhaps it is possible to teach through the network of the TV set with a fridge and an iron)
At this point, perhaps, we can safely enter the evolutionary dead end and give way to other normal species :D
In general, it is necessary to evaluate whether those advantages of ONNX application are worth the effort to learn and use it. This is for everyone to decide.
Exactly what I've said many times here--
no one has any working model, but everyone needs to know how to implement a deep model via onnx :)
Exactly what I've said many, many times here.
no one has any working model, but everyone needs to know how to implement a deep model via onnx :)
Well, at least out of respect for the work done by the developers it is necessary to feel and study the proposed feature. And to apply or not to apply let everyone decide for himself. Besides, today it is complicated and incomprehensible, and tomorrow there may be an opportunity to simplify this conversion process.
You phrase it wrong: "No one has a working model..." I would say that the forum doesn't see many models actually working in the market. But that's fine. Nobody is going to give away their hard-earned money for nothing.
Good luck
Well, at least out of respect for the work done by the developers it is necessary to feel, to study the proposed opportunity. And let everyone decide for himself whether to use it or not. Besides, today it is complicated and incomprehensible, and tomorrow it may be possible to simplify the conversion process.
Yes of course what is done is fine ) and will be useful ...
It is just necessary to follow some logical sequence of actions and thoughts ....
to spend a lot of resources and time on learning an important and necessary technology to launch a model WHICH ISN'T?!?! is not consistent....
You are incorrectly formulating: "Nobody has a working model..."
I suspected that I would be corrected....
Yes, of course I meant the vast majority > 95~99.9 %.
Exactly what I've said many, many times here.
no one has any working model, but everyone needs to know how to implement a deep model via onnx :)
I have the impression that NO ONE has a full-blooded MM EA that would use the model for prediction, at least at the tester level.
For example, I have such an Expert Advisor with a very good prediction model (prediction error is less than 20%), but an extremely unpleasant peculiarity has appeared: half of the prediction error, i.e. 10% error, gives about the same loss as all other positive entries.
Rhetorical question: will ONNX help to solve this problem?
By the way, I have been saying the same thing for all three thousand pages: there is no problem with models, using any of the many hundreds of models is the easiest thing to do when developing MO-based Expert Advisors.