search

assume-framework / assume

ASSUME - Agent-based Simulation for Studying and Understanding Market Evolution
https://assume.readthedocs.io
20 stars 5 forks source link

Adjust observations for learning agents for best learning performance and speed #339

Open kim-mskw opened 2 months ago

kim-mskw commented 2 months ago

In the initial implementation of RL we disregarede the historic values in the observation space metioned in Harder et al. These need to be added back to ensure good performance for larger simulations.

nick-harder commented 2 weeks ago

As mentioned by @kim-mskw during a development meeting, reducing the observation space has also improved the performance. We should investigate this as part of the work