I am trying to implement DDPG in a custom gym environment. The action is the relative allocation of funds between each asset. The action space is a Box with the shape of the available assets, with a minimum of 0 and a maximum of 1. Therefore, I need to make sure that the total sum of the actions $a_1, a_2,...,a_N$ equals one: $\sum_{i=1}^N a_i= 1$. So far, I have found two different approaches:

  1. Braithwaite scales the actions inside the customized environment: https://github.com/acbraith/gym-asset_allocation

  2. Fontura et al. scale the actions inside the DDPG algorithm: https://github.com/MLRG-CEFET-RJ/DRL-ALM (spinup -> algos -> pytorch -> ddpg> -> ddpg.py)

Which approach is recommended? And how can the different approaches affect the performance of the DDPG?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.