Machine learning in trading: theory, models, practice and algo-trading - page 3386

New comment
 
Maxim Dmitrievsky #:
If the error has already stopped dropping or is equal to zero, you can divide the remaining examples into patterns by some measure of closeness :) Clustering, for example. And count how many are left. And even write an averaged condition for each pattern/cluster (take centroids of clusters), you will get a rule in the output.
It can be done, but where is the guarantee that the partitioning was done correctly, in accordance with the internal partitioning of the neuronics?

Where is the guarantee that the chosen proximity measure for clustering is chosen correctly?

Etc...

Wouldn't it be simpler to partition the wooden model and not have to create a Franklinstein

 
mytarmailS #:
It can be done, but where is the guarantee that the partitioning is correct, according to the internal partitioning of the neuronics?
Where is the guarantee that the chosen proximity measure for clustering is chosen correctly?
Well, that's a philosophical question.
 
This is where the last layer of neuronics, not the examples, should be clustered, if not the examples
 
mytarmailS #:
If you're going to cluster, it's not the examples, it's the last layer of neuronics.
No, it's the examples. You can't build rules on the last layers of neuronics.
 
Maxim Dmitrievsky #:
No, just examples. You can't build rules from the last layers of neuronics.
Let me create a sample data and everyone apply their own methodology and we'll see.
 
mytarmailS #:
Let me create a sample data and everyone apply their methodology and we'll see
I haven't tried this approach, just thinking out loud about how to get rules out of any model. We can play around with this later.
 
Maxim Dmitrievsky #:
I haven't tried this approach, just thinking out loud about how to get rules out of any model. You can play around with it later.
I haven't tried it either, it's my theory against yours
 

It seems that the articles have stopped being translated or they don't have time. The English section is already full of python articles and onnx :)) And one article on R has appeared.

In general, the articles are useless in terms of TC. About the same as on Medium, they write to write.

 
mytarmailS #:

Now identify all linearly related rules and remove them as redundant rules

How is this linear coupling defined? Can you elaborate?

I just remove rules that are very similar, similarity is determined by activation points.

 
Aleksey Vyazmikin #:

How is this linear relationship defined? Can you elaborate?

I just remove rules that are very similar, I determine similarity by activation points.

I published the code.
Here are the details


What other activation points?
1...337933803381338233833384338533863387338833893390339133923393...3512
New comment
Reason: