V0.6.1: Final version without wall observations

Depreciation Note

This is the final version that is still compatible in terms of tensor shapes for state observations with models trained as described in Gym-μRTS paper.

After this version, we are going to add an extra channel to state observations representing the terrain (whether or not there is a wall on any given cell), which means that models trained for the old code will be incompatible because they'll be receiving larger input tensors than they expect.

What's Changed

• Minor documentation changes by @DennisSoemers in #105, #123, #126
• Allow passing custom JVM args by @DennisSoemers in #109
• Minor optimisations by @DennisSoemers in #111
• Updated unit testing by @DennisSoemers in #124

Full Changelog: v0.6.0...v0.6.1

V0.6.0 Speed improvement; better league evaluation; entity gym support

What's Changed

• Multiple peformance improvements for vec_env by @kachayev in #30
• Add test cases by @vwxyzjn in #31
• Simplify mask implementation by @vwxyzjn in #32
• Better league convergence criterion by @vwxyzjn in #37
• Zero-copy zero-allocation shared buffer between JVM and NumPy arrays by @kachayev in #34
• Prototype quality-based match making by @vwxyzjn in #43
• Add observation and mask test cases by @vwxyzjn in #38
• Refactor league and produce better documentation by @vwxyzjn in #44
• Use submodules for static files by @vwxyzjn in #45
• Remove historic scripts by @vwxyzjn in #47
• Add end-to-end test cases by @vwxyzjn in #48
• Refactor eval script by @vwxyzjn in #49
• Add pre-commit utilities by @vwxyzjn in #46
• Log discounted returns by @vwxyzjn in #52
• Improve league DB cache and API by @vwxyzjn in #53
• Training evaluation improvement by @vwxyzjn in #54
• Introduce smaller and more efficient NN architecture by @vwxyzjn in #56
• Cosmetic updates to gridnet script by @kachayev in #62
• Transfer multiple maps support by @xluox in #66
• Prototype entity gym support by @vwxyzjn in #69

Full Changelog: v0.5.0...v0.6.0

V0.5.0 New Action Space Definition

What's Changed

• Added evalucation process for PO by @xluox in #14
• Put Competition Bots in libs by @vwxyzjn in #15
• Use the correct action space definition by @vwxyzjn in #16
• Support for passing different maps in envs by @vwxyzjn in #17
• Make the spyder dependencies optional by @vwxyzjn in #18
• Enable users to evaluate two different agent models by @vwxyzjn in #19
• Allow evaluate two rl agents using Trueskill by @vwxyzjn in #20
• Add Trueskill Evaluation for New AIs by @vwxyzjn in #21
• Quick bug fix with league evaluation by @vwxyzjn in #23
• Add docker and cloud integration by @vwxyzjn in #22
• Set a default league database by @vwxyzjn in #24
• Fix wandb tensorboard bug of not logging metrics by @vwxyzjn in #25
• Hotfix: Log Trueskill again by @vwxyzjn in #26
• Better visualization by @vwxyzjn in #27
• Add partial observable league.db by @xluox in #28

New Contributors

• @xluox made their first contribution in #14

Full Changelog: v0.4.3...v0.5.0

v0.4.3

Merge pull request #13 from vwxyzjn/new-action-space

New action space

v0.4.2

update dependencies

v0.4.1

Remove stable-baselines3 as an essential dependency

v0.4.0

update

v0.3.5

update poetry config

v0.3.4

update gridnet example

v0.3.3

restore flow