- enable atari env to use full action set of 18
- modify rendering.py to prevent some bad stack traces upon closing
- enable dict space to be more concisely specified as spaces.Dict(a=Box(…),…)
* add dtype to Box
* remove board_game, debugging, safety, parameter_tuning environments
* massive set of breaking changes
- remove python logging module
- _step, _reset, _seed, _close => non underscored method
- remove benchmark and scoring folder
* Improve render("human"), now resizable, closable window.
* get rid of default step and reset in wrappers, so it doesn’t silently fail for people with underscore methods
* CubeCrash unit test environment
* followup fixes
* MemorizeDigits unit test envrionment
* refactored spaces a bit
fixed indentation
disabled test_env_semantics
* fix unit tests
* fixes
* CubeCrash, MemorizeDigits tested
* gym backwards compatibility patch
* gym backwards compatibility, followup fixes
* changelist, add spaces to main namespaces
* undo_logger_setup for backwards compat
* remove configuration.py
save (clone) and load (restore) the Atari emulator state for analysis,
planning, or other experiments that require returning to identical
states of the environment.
note: `clone_full_state()` and `restore_full_state()` are necessary to
save and load the ROM and system state (including pseudorandomness) for
fully identical restoration of the environment.
Get screen pixels in a way that doesn’t require swapping the RGB
channels.
SeaquestNoFrameskip used to take 3.5 seconds to do 10k steps, now it
takes 2.6 seconds.
When also using opencv to resize to (84,84,3), it used to take 7.8
seconds and now takes 7.0.
* fix double reset, as suggested by @jietang
* better floors and ceilings
* add convenience methods to monitor
* add wrappers to gym namespace
* allow playing Atari games, with potentially more coming in the future
* simplify example in docs
* Move play out of the Env
* fix tests
* no more deprecation warnings
* remove env.monitor
* monitor simplification
* monitor simplifications
* monitor related fixes
* a few changes suggested by linter
* timestep_limit fixes
* keep track of gym env variables for future compatibility
* timestep_limit => max_episode_timesteps
* don't apply TimeLimit wrapper in make for VNC envs
* Respect old timestep_limit argument
* Pass max_episode_seconds through registration
* Don't include deprecation warnings yet
We intended to use this functionality in Universe, but have
decided to keep it within the Universe repo, and expose
ALE lives in the info dict instead (see next commit)
* Make environments seedable
* Fix monitor bugs
- Set monitor_id before setting the infix. This was a bug that would yield incorrect results with multiple monitors.
- Remove extra pid from stats recorder filename. This should be purely cosmetic.
* Start uploading seeds in episode_batch
* Fix _bigint_from_bytes for python3
* Set seed explicitly in random_agent
* Pass through seed argument
* Also pass through random state to spaces
* Pass random state into the observation/action spaces
* Make all _seed methods return the list of used seeds
* Switch over to np.random where possible
* Start hashing seeds, and also seed doom engine
* Fixup seeding determinism in many cases
* Seed before loading the ROM
* Make seeding more Python3 friendly
* Make the MuJoCo skipping a bit more forgiving
* Remove debugging PDB calls
* Make setInt argument into raw bytes
* Validate and upload seeds
* Skip box2d
* Make seeds smaller, and change representation of seeds in upload
* Handle long seeds
* Fix RandomAgent example to be deterministic
* Handle integer types correctly in Python2 and Python3
* Try caching pip
* Try adding swap
* Add df and free calls
* Bump swap
* Bump swap size
* Try setting overcommit
* Try other sysctls
* Try fixing overcommit
* Try just setting overcommit_memory=1
* Add explanatory comment
* Add what's new section to readme
* BUG: Mark ElevatorAction-ram-v0 as non-deterministic for now
* Document seed
* Move nondetermistic check into spec
