Discussion of article "Gradient boosting in transductive and active machine learning"

New comment
 

New article Gradient boosting in transductive and active machine learning has been published:

In this article, we will consider active machine learning methods utilizing real data, as well discuss their pros and cons. Perhaps you will find these methods useful and will include them in your arsenal of machine learning models. Transduction was introduced by Vladimir Vapnik, who is the co-inventor of the Support-Vector Machine (SVM).

Let us go straight to active learning and test its effectiveness on our data.

There are several libraries for active learning in the Python language, the most popular of them being:

  • modAL is quite a simple and easy-to-learn package, which is a kind of a wrapper for the popular machine learning library scikit-learn (they are fully compatible). The package provides the most popular active learning methods.
  • Libact uses the multi-armed bandit strategy over existing query strategies for a dynamic selection of the best query. 
  • Alipy is a kind of a laboratory from package providers, which contains a large number of query strategies.

I have selected the modAL library as being more intuitive and suitable for getting acquainted with the active learning philosophy. It offers greater freedom in designing models and in creating your own models by using standard blocks or by creating your own ones.

Let us consider the above described process using the below scheme, which does not require further explanations:

Author: Maxim Dmitrievsky

 
MetaQuotes:

Published paper Gradient bousting in transductive and active machine learning tasks:

Author: Maxim Dmitrievsky

Norm, of course the presence of manual labels and reasonable assumptions is not AI) but really reduces the area of data and dimensions for training, makes them acceptable at least).

[Deleted]  
Valeriy Yastremskiy:

Norm, of course having hand-tags and reasonable assumptions is not AI) but really reduces the data area and dimensions for training, makes them acceptable at least)

From manual only selection of features (fiches). I'm thinking how to get rid of the last manual.

There are options, but I haven't got my hands on it yet
 
Maxim Dmitrievsky:

From the manual only selection of features (features). I'm thinking how to get rid of the last manual.

There are options, but I haven't got my hands on it yet.

And automatic detection of targets))))

[Deleted]  
Valeriy Yastremskiy:

And automatic targeting))))

this is already in place

 
Thank you for the wonderful article.
[Deleted]  
Por:
Thank you for the wonderful article.

You are welcome.

 

I don't understand anything :)

How many classes were marked up automatically?

If more than two, it would be logical to give each class a financial assessment - whether it makes a loss or profit, and then merge into two classes for final training.

[Deleted]  
Aleksey Vyazmikin:

Didn't understand a thing :)

How many classes were marked up automatically?

If more than two, it would be logical to give each class a financial evaluation - whether it makes a loss or profit, and then merge into two classes for final training.

What can I do to alleviate the misunderstanding? 2 classes, as usual

The question of options for applying active learning is open. This is too big and, in part, philosophical and experimental to do all in one article
 

Congratulations for such a great article after a long time!!!


How to train and test other currency pairs?


The coding part is complicated for me to do any editing or any form of improvements for testing purpose:)


Can you please help with the errors in the screenshot?

Thanks
Files:
USDCAD_pair.png  46 kb
 
FxTrader562:

Congratulations for such a great article after a long time!!!


How to train and test other currency pairs?


The coding part is complicated for me to do any editing or any form of improvements for testing purpose:)


Can you please help with the errors in the screenshot?

Thanks

You can not compile py file in MetaEditor5, for this purpose you need to install PyCharm program and run this script.

123
New comment