v1.5.2

[jsuarez5341]

v1.5.2: Ray Tune and WanDB Integrations

• Trinity:
  • Added support for simulations with both scripted and trained agents
  • Added ability to name scripted agents based on their policy
• Embyr:
  • Minor aesthetic changes to prefer a flat-shaded style
  • Broke some overlay features :/ RLlib bug under construction
• /projekt
  • Replaced Bokeh dashboard with WanDB integration
  • Wrapped RLlib trainers in Ray Tune to enable parallel evaluation during training
  • Added Skill Rating (SR) metric for direct comparison to scripted baselines
  • Changed batching mode to agent steps, yielding a large policy improvement