Move add_experiences out of trainer, add Trajectories#3067

Merged

ervteng merged 79 commits intomasterfrom

develop-agentprocessor

Dec 19, 2019

Commits on Nov 22, 2019

Split buffer into two buffers (PPO works)
Ervin Teng
committed

Commits on Nov 25, 2019

buffer split for SAC
Ervin Teng
committed
Fix buffer tests and truncate
Ervin Teng
committed
Fix RL tests
Ervin Teng
committed
Fix demo loader and remaining tests
Ervin Teng
committed
Remove MANIFEST file
Ervin Teng
committed
Add type hints to Buffer
Ervin Teng
committed
Rename append_update_buffer to append_to_update_buffer
Ervin Teng
committed

Commits on Nov 26, 2019

Merge branch 'develop' into develop-splitbuffer
Ervin Teng
committed
Non-working commit
Ervin Teng
committed
Revert buffer for now
Ervin Teng
committed

Commits on Nov 27, 2019

Another nonworking commit
Ervin Teng
committed
Runs but doesn't do anything yet
Ervin Teng
committed

Commits on Dec 4, 2019

Merge branch 'develop' into develop-agentprocessor
Ervin Teng
committed
Use ProcessingBuffer in AgentProcessor
Ervin Teng
committed
Convert to trajectory
Ervin Teng
committed

Commits on Dec 5, 2019

Looks like it's training
Ervin Teng
committed
Fix memory leak
Ervin Teng
committed
Attempt reward reporting
Ervin Teng
committed
Stats reporting is working
Ervin Teng
committed
Clean up some stuff
Ervin Teng
committed
No longer using ProcessingBuffer for PPO
Ervin Teng
committed
Move trajectory and related functions to trajectory.py
Ervin Teng
committed

Commits on Dec 6, 2019

Add back max_step logic
Ervin Teng
committed
Merge branch 'master' of github.com:Unity-Technologies/ml-agents into develop-agentprocessor
Ervin Teng
committed
Remove epsilon
Ervin Teng
committed
Migrate SAC
Ervin Teng
committed
Remove dead code
Ervin Teng
committed
Move some common logic to buffer class
Ervin Teng
committed
Kill the ProcessingBuffer
Ervin Teng
committed
Convert BC (warning) might be broken
Ervin Teng
committed
Fix some bugs for visual obs
Ervin Teng
committed

Commits on Dec 7, 2019

Fixes for recurrent
Ervin Teng
committed
Better decoupling for agent processor
Ervin Teng
committed
Fix some of the tests
Ervin Teng
committed

Commits on Dec 9, 2019

Add test for trajectory
Ervin Teng
committed
Fix BC and tests
Ervin Teng
committed
Lots of test fixes
Ervin Teng
committed
Remove BootstrapExperience
Ervin Teng
committed
Move agent_id to Trajectory
Ervin Teng
committed

Commits on Dec 10, 2019

Add back next_obs
Ervin Teng
committed
Fix test again
Ervin Teng
committed
Fix PPO value tests
Ervin Teng
committed
Properly report value estimates and episode length
Ervin Teng
committed
Fix np float32 errors
Ervin Teng
committed
Fix one more np float32 issue
Ervin Teng
committed
Merge branch 'master' into develop-agentprocessor
Ervin Teng
committed
Fix some import errors
Ervin Teng
committed

Commits on Dec 11, 2019

Make conversion methods part of NamedTuples
Ervin Teng
committed
Add way to check if trajectory is done or max_reached
Ervin Teng
committed
Add docstring
Ervin Teng
committed
Address AgentProcessor comments
Ervin Teng
committed

Commits on Dec 12, 2019

Allow None max steps
Ervin Teng
committed
Merge branch 'master' into develop-agentprocessor
Ervin Teng
committed
Fix tests
Ervin Teng
committed
Fix some mypy issues and remove unused code
Ervin Teng
committed
Fix numpy import
Ervin Teng
committed
Remove defaultdict that didn't make sense
Ervin Teng
committed
Fixed value estimate bug
Ervin Teng
committed
Fix mypy issue
Ervin Teng
committed

Commits on Dec 13, 2019

Add stats reporter class and re-enable missing stats (#3076)
Ervin T
authored
Revert gitignore
Ervin Teng
committed

Commits on Dec 14, 2019

Normalize based on number of elements
Ervin Teng
committed

Commits on Dec 16, 2019

Add comment
Ervin Teng
committed
Merge branch 'master' into develop-agentprocessor
Ervin Teng
committed

Commits on Dec 17, 2019

New way to update mean and var
Ervin Teng
committed

Commits on Dec 18, 2019

Merge branch 'master' into develop-agentprocessor
Ervin Teng
committed
Fix tests
Ervin Teng
committed
Add comments for normalization
Ervin Teng
committed
Remove dead code
Ervin Teng
committed

Commits on Dec 19, 2019

Add type hints to rl_trainer
Ervin Teng
committed
Cleanup agent_processor
Ervin Teng
committed
Make file creation safer
Ervin Teng
committed
Fix error message
Ervin Teng
committed
Clean up trajectory and splitobs
Ervin Teng
committed
Use .get for trainer_parameters
Ervin Teng
committed
Add test for normalization
Ervin Teng
committed
Float32 array in test
Ervin Teng
committed
Fix comment in test
Ervin Teng
committed