8000 Tags · koulanurag/ma-gym · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

Tags: koulanurag/ma-gym

Tags

v0.0.14

Toggle v0.0.14's commit message
chore: bump version

v0.0.13

Toggle v0.0.13's commit message
build: bump version

v0.0.11

Toggle v0.0.11's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
Merge pull request #46 from dongchirua/master

- package version updated
- fix minor bug

v0.0.10

Toggle v0.0.10's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
fix: update pyglet (#45)

fix error when rendering environment in a server 
the error: `xvfb-run -s "-screen 0 1400x900x24"`

10000 v0.0.9

Toggle v0.0.9's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
build: bump up version

v0.0.8

Toggle v0.0.8's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
fix: combat (#41)

Fix bugs:
    1. in get_agent_obs(), the local variable '_type' was mistakenly written as 'type'.
    2. self.agent_prev_pos and self.opp_prev_pos are not updated, but in __update_agent_view() and __update_opp_view(), these previous positions are cleared. Thus the movement records are incorrect.
    3. The self._agent_cool and self._opp_cool are not updated (i.e., the agents and opponents could execute 'attack' action at every step if within range).
        updates:
          (1) update the is_fireable() function to check if the attack agents have been cooled down. 
          (2) add self._agent_cool_step and self._opp_cool_step to track the cooling down of the agents and opponents.
    4. In the opps_action() function, the dead opponent could still call reduce_distance_move() function to sample a 'move' action, which will raise an 'no where to move' exception when the position of the dead opponent is the same with any of the agents.
    5. Update the get_agent_obs(): only return the local view of observation when the agent is alive, otherwise return a zero vector.

Add 2 new functions to support CTDE
    6. Add get_state() to support CTDE (centralized training with decentralized execution).

In get_agent_obs(), the x-coordinates and y-coordinates of the entites (whether the entity is 'agent' or 'opponent') within each agent's view range were incorrectly set to the agent's position.

v0.0.7

Toggle v0.0.7's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
feat: add clock versions (#35)

- switch and checkers has v2 and v3 version with clock info.

v0.0.6

Toggle v0.0.6's commit message
chore: bump version

- version includes small fixes.

v0.0.5

Toggle v0.0.5's commit message
bump version

v0.0.4

Toggle v0.0.4's commit message

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
workflows added (#25)

- test and publish
- bump version
0