SAC (both torch and tf versions) are showing issues (crashes) due to numeric instabilities in the SquashedGaussian distribution (sampling + logp after extreme NN outputs).
This PR fixes these. Stable MuJoCo learning (HalfCheetah) has been confirmed on both tf and torch versions. A Distribution stability test (using extreme NN outputs) has been added for SquashedGaussian (can be used for any other type of distribution as well).
* Policy-classes cleanup and torch/tf unification.
- Make Policy abstract.
- Add `action_dist` to call to `extra_action_out_fn` (necessary for PPO torch).
- Move some methods and vars to base Policy
(from TFPolicy): num_state_tensors, ACTION_PROB, ACTION_LOGP and some more.
* Fix `clip_action` import from Policy (should probably be moved into utils altogether).
* - Move `is_recurrent()` and `num_state_tensors()` into TFPolicy (from DynamicTFPolicy).
- Add config to all Policy c'tor calls (as 3rd arg after obs and action spaces).
* Add `config` to c'tor call to TFPolicy.
* Add missing `config` to c'tor call to TFPolicy in marvil_policy.py.
* Fix test_rollout_worker.py::MockPolicy and BadPolicy classes (Policy base class is now abstract).
* Fix LINT errors in Policy classes.
* Implement StatefulPolicy abstract methods in test cases: test_multi_agent_env.py.
* policy.py LINT errors.
* Create a simple TestPolicy to sub-class from when testing Policies (reduces code in some test cases).
* policy.py
- Remove abstractmethod from `apply_gradients` and `compute_gradients` (these are not required iff `learn_on_batch` implemented).
- Fix docstring of `num_state_tensors`.
* Make QMIX torch Policy a child of TorchPolicy (instead of Policy).
* QMixPolicy add empty implementations of abstract Policy methods.
* Store Policy's config in self.config in base Policy c'tor.
* - Make only compute_actions in base Policy's an abstractmethod and provide pass
implementation to all other methods if not defined.
- Fix state_batches=None (most Policies don't have internal states).
* Cartpole tf learning.
* Cartpole tf AND torch learning (in ~ same ts).
* Cartpole tf AND torch learning (in ~ same ts). 2
* Cartpole tf (torch syntax-broken) learning (in ~ same ts). 3
* Cartpole tf AND torch learning (in ~ same ts). 4
* Cartpole tf AND torch learning (in ~ same ts). 5
* Cartpole tf AND torch learning (in ~ same ts). 6
* Cartpole tf AND torch learning (in ~ same ts). Pendulum tf learning.
* WIP.
* WIP.
* SAC torch learning Pendulum.
* WIP.
* SAC torch and tf learning Pendulum and Cartpole after cleanup.
* WIP.
* LINT.
* LINT.
* SAC: Move policy.target_model to policy.device as well.
* Fixes and cleanup.
* Fix data-format of tf keras Conv2d layers (broken for some tf-versions which have data_format="channels_first" as default).
* Fixes and LINT.
* Fixes and LINT.
* Fix and LINT.
* WIP.
* Test fixes and LINT.
* Fixes and LINT.
Co-authored-by: Sven Mika <sven@Svens-MacBook-Pro.local>
* Exploration API (+EpsilonGreedy sub-class).
* Exploration API (+EpsilonGreedy sub-class).
* Cleanup/LINT.
* Add `deterministic` to generic Trainer config (NOTE: this is still ignored by most Agents).
* Add `error` option to deprecation_warning().
* WIP.
* Bug fix: Get exploration-info for tf framework.
Bug fix: Properly deprecate some DQN config keys.
* WIP.
* LINT.
* WIP.
* Split PerWorkerEpsilonGreedy out of EpsilonGreedy.
Docstrings.
* Fix bug in sampler.py in case Policy has self.exploration = None
* Update rllib/agents/dqn/dqn.py
Co-Authored-By: Eric Liang <ekhliang@gmail.com>
* WIP.
* Update rllib/agents/trainer.py
Co-Authored-By: Eric Liang <ekhliang@gmail.com>
* WIP.
* Change requests.
* LINT
* In tune/utils/util.py::deep_update() Only keep deep_updat'ing if both original and value are dicts. If value is not a dict, set
* Completely obsolete syn_replay_optimizer.py's parameters schedule_max_timesteps AND beta_annealing_fraction (replaced with prioritized_replay_beta_annealing_timesteps).
* Update rllib/evaluation/worker_set.py
Co-Authored-By: Eric Liang <ekhliang@gmail.com>
* Review fixes.
* Fix default value for DQN's exploration spec.
* LINT
* Fix recursion bug (wrong parent c'tor).
* Do not pass timestep to get_exploration_info.
* Update tf_policy.py
* Fix some remaining issues with test cases and remove more deprecated DQN/APEX exploration configs.
* Bug fix tf-action-dist
* DDPG incompatibility bug fix with new DQN exploration handling (which is imported by DDPG).
* Switch off exploration when getting action probs from off-policy-estimator's policy.
* LINT
* Fix test_checkpoint_restore.py.
* Deprecate all SAC exploration (unused) configs.
* Properly use `model.last_output()` everywhere. Instead of `model._last_output`.
* WIP.
* Take out set_epsilon from multi-agent-env test (not needed, decays anyway).
* WIP.
* Trigger re-test (flaky checkpoint-restore test).
* WIP.
* WIP.
* Add test case for deterministic action sampling in PPO.
* bug fix.
* Added deterministic test cases for different Agents.
* Fix problem with TupleActions in dynamic-tf-policy.
* Separate supported_spaces tests so they can be run separately for easier debugging.
* LINT.
* Fix autoregressive_action_dist.py test case.
* Re-test.
* Fix.
* Remove duplicate py_test rule from bazel.
* LINT.
* WIP.
* WIP.
* SAC fix.
* SAC fix.
* WIP.
* WIP.
* WIP.
* FIX 2 examples tests.
* WIP.
* WIP.
* WIP.
* WIP.
* WIP.
* Fix.
* LINT.
* Renamed test file.
* WIP.
* Add unittest.main.
* Make action_dist_class mandatory.
* fix
* FIX.
* WIP.
* WIP.
* Fix.
* Fix.
* Fix explorations test case (contextlib cannot find its own nullcontext??).
* Force torch to be installed for QMIX.
* LINT.
* Fix determine_tests_to_run.py.
* Fix determine_tests_to_run.py.
* WIP
* Add Random exploration component to tests (fixed issue with "static-graph randomness" via py_function).
* Add Random exploration component to tests (fixed issue with "static-graph randomness" via py_function).
* Rename some stuff.
* Rename some stuff.
* WIP.
* WIP.
* Fix SAC.
* Fix SAC.
* Fix strange tf-error in ray core tests.
* Fix strange ray-core tf-error in test_memory_scheduling test case.
* Fix test_io.py.
* LINT.
* Update SAC yaml files' config.
Co-authored-by: Eric Liang <ekhliang@gmail.com>
