Discussing the article: "Neural networks made easy (Part 62): Using Decision Transformer in hierarchical models"

[MetaQuotes]

Check out the new article: Neural networks made easy (Part 62): Using Decision Transformer in hierarchical models.

In recent articles, we have seen several options for using the Decision Transformer method. The method allows analyzing not only the current state, but also the trajectory of previous states and actions performed in them. In this article, we will focus on using this method in hierarchical models.

Collecting a training sample for the historical period in the first 7 months of 2023 turned out to be quite labor-intensive. I ran into the problem that even with a small sampling horizon of Agent actions, most passes did not satisfy the positive balance requirement.

 

To select the optimal planning horizon in the optimization mode, the number of iterations per pass was adjusted to the optimized parameters.

After collecting the training set and training the local policy model, I ran the scheduler and cost function model training in parallel. This approach allowed me to significantly reduce the time spent training models.

Author: Dmitriy Gizlyk

[Fahad Raza]

It aims to foster a deeper understanding of Decision Transformers in hierarchical architectures, particularly for those interested in its use cases for robotics and autonomous systems.