Releases: davidtjones/melee-env
Releases · davidtjones/melee-env
0.30.0
Agent framework overhaul
Melee-env is now more inclusive, with support for both learning and non-learning agents. Other improvements:
- The agent framework has also been streamlined
- 2p hud alignment is automatic
- bugfixes
- documentation updates
- tmp home directory support across platforms
0.22.1
Windows support, new features, bugfixes
- Windows support 🎉
*All agents now consume observations and the action space so that the
game knows when they have been defeated viais_defeated
.
- Agents store
self_observation
. This is inserted instead of [-1, -1,
-1, -1...] into the observation when agents are defeated. The previous
design was problematic for agents who utilized position. The new method
preserves the agent's final state, which should proivide indication
about whether that opponent is still targetable. - Dynamic toggle for 2P Center Aligned HUD
- Rest agent no longer chases defeated opponents.
- README updates
0.21.0
More refactoring, another basic agent type, docs
- Made
'AI'
the default agent type. No longer needs to be defined in
Agent subclasses. - New NOOP agent that selects its character and then does nothing.
- Removed some unused code in util.py
0.20.2
New Rest agent along with other fixes and changes.
- Observation spaces now require stocks to be in the last channel
- updated the example observation space to support this new requirement and to also contain player positional data
- implemented AltF4's puff bot from his tutorial video using the example action space
- redesigned is_defeated as a function decorator to reduce boilerplate code on new agents
- updated
- updated documentation in agents/README.md
- updated example file