1
0
mirror of https://github.com/gryf/coach.git synced 2025-12-18 03:30:19 +01:00

Commit Graph

  • 4a6c404070 Adding worker logs and plumbed task_parameters to distributed coach (#130) Ajay Deshpande 2018-11-23 15:35:11 -08:00
  • 2b4c9c6774 Removing grarph_manager param (#141) Gal Leibovich 2018-11-23 21:42:54 +02:00
  • a1c56edd98 Fixes for having NumpySharedRunningStats syncing on multi-node (#139) Gal Leibovich 2018-11-23 16:11:47 +02:00
  • 87a7848b0a Moved tf.variable_scope and tf.device calls to framework-specific architecture (#136) Sina Afrooze 2018-11-22 12:52:22 -08:00
  • 559969d3dd disabled loading for target weights (#138) shadiendrawis 2018-11-22 18:15:52 +02:00
  • 949d91321a Added explicit environment closing (#129) Thom Lane 2018-11-22 04:25:03 -08:00
  • 16cdd9a9c1 Tf checkpointing using saver mechanism (#134) Sina Afrooze 2018-11-22 04:08:10 -08:00
  • dd18959e53 Don't download when checkpoint files are already present (#109) Cody Hsieh 2018-11-21 15:32:53 -08:00
  • b94239234a Removed TF warning when training in a distributed setting (#133) shadiendrawis 2018-11-21 16:09:04 +02:00
  • a112ee69f6 Save filters' internal state (#127) Gal Leibovich 2018-11-20 17:21:48 +02:00
  • 67eb9e4c28 Adding checkpointing framework (#74) Sina Afrooze 2018-11-19 09:45:49 -08:00
  • 4da56b1ff2 Enable setting the data store factory in Graph manager (#110) x77a1 2018-11-19 08:35:03 -08:00
  • 67a90ee87e Add tensor input type for arbitrary dimensional observation (#125) Sina Afrooze 2018-11-19 06:41:12 -08:00
  • 7ba1a4393f Channel order transpose, for image embedder. Updated unit test. (#87) Thom Lane 2018-11-19 05:39:03 -08:00
  • ff816b347d aws pip package (#118) shadiendrawis 2018-11-19 14:00:16 +02:00
  • 3817cefb12 removing box2d and atari requirements (#124) Gal Novik 2018-11-19 13:42:08 +02:00
  • 9210909050 Added MXNet to arg docs. (#121) Thom Lane 2018-11-19 01:31:28 -08:00
  • d4d06aaea6 remove kubernetes dependency (#117) Gal Leibovich 2018-11-18 18:10:22 +02:00
  • 430e286c56 muting pygame's hello message (#116) Gal Leibovich 2018-11-18 18:02:55 +02:00
  • ce85c8e8c3 Removing Egreedy from CartPole_ClippedPPO. ClippedPPO's default exploration policy is to be used instead. (#115) Gal Leibovich 2018-11-18 16:36:34 +02:00
  • 6caf721d1c Numpy shared running stats (#97) Gal Leibovich 2018-11-18 14:46:40 +02:00
  • e1fa6e9681 roboschool: updating envs to v1, fixing rendering (#112) Gal Novik 2018-11-18 13:38:10 +02:00
  • 9fd4d55623 Making stop condition optional by using a flag (#113) Gal Leibovich 2018-11-18 13:37:39 +02:00
  • 449bcfb4e1 summing head losses instead of taking the mean (#98) Gal Leibovich 2018-11-18 12:20:00 +02:00
  • 5b11fa5656 check for local mujoco key in build process (#105) Zach Dwiel 2018-11-18 03:57:30 -05:00
  • dea1826658 Re-enable NFS data store. (#101) Balaji Subramaniam 2018-11-16 13:55:33 -08:00
  • a0f25034c3 Added average total reward to logging after evaluation phase completes. (#93) Thom Lane 2018-11-16 08:22:00 -08:00
  • 81bac050d7 Added Custom Initialisation for MXNet Heads (#86) Thom Lane 2018-11-16 08:15:43 -08:00
  • 101c55d37d Handle both Environment Steps and Episodes on the subscriber side. (#99) Balaji Subramaniam 2018-11-15 14:42:21 -08:00
  • 3358e04a6a Corrected MXNet's PPO Head for Continuous Action Spaces (#84) Thom Lane 2018-11-15 13:27:54 -08:00
  • fde73ced13 Simulating the act on the trainer. (#65) Ajay Deshpande 2018-11-15 08:38:58 -08:00
  • fe6857eabd broaden supported package versions (#50) Scott Leishman 2018-11-15 05:29:49 -08:00
  • 6d40ad1650 update of api docstrings across coach and tutorials [WIP] (#91) Itai Caspi 2018-11-15 15:00:13 +02:00
  • 524f8436a2 create per environment Dockerfiles. (#70) Scott Leishman 2018-11-14 07:40:22 -08:00
  • a849c17e46 Enable distributed SharedRunningStats (#81) Balaji Subramaniam 2018-11-13 09:17:38 -08:00
  • 875d6ef017 Adding target reward and target sucess (#58) Ajay Deshpande 2018-11-12 15:03:43 -08:00
  • 0fe583186e fixing the coach entrypoint after adding the CoachLauncher abstraction (#92) Itai Caspi 2018-11-12 20:26:49 +02:00
  • 2804a7c24f Refactor launcher to be object-oriented (#63) Leo Dirac 2018-11-10 12:10:19 -08:00
  • 3fd433ffab fix ddpg head (#78) Itai Caspi 2018-11-09 18:17:04 +02:00
  • 3a0a1159e9 fixing the dropout rate code (#72) Itai Caspi 2018-11-08 16:53:47 +02:00
  • 389c65cbbe fix for a bug in distributed training that was introduced lately (#75) Itai Caspi 2018-11-08 16:52:48 +02:00
  • 83e0b09a6a adding the missing export_onnx_graph parameter to task parameters (#73) Itai Caspi 2018-11-08 12:52:42 +02:00
  • 8f0415b4cc Tweak additional_simulator_parameters for easier configuration and better error logging. (#69) Leo Dirac 2018-11-07 11:01:12 -08:00
  • 49dea39d34 N-step returns for rainbow (#67) Gal Leibovich 2018-11-07 18:33:08 +02:00
  • 35c477c922 allowing grayscale observations in gym (#66) Itai Caspi 2018-11-07 17:08:10 +02:00
  • 5fadb9c18e Adding mxnet components to rl_coach/architectures (#60) Sina Afrooze 2018-11-07 07:07:15 -08:00
  • e7a91b4dc3 Fix cmd line arguments handling (#68) Itai Caspi 2018-11-07 15:47:02 +02:00
  • 93571306c3 Removed tensorflow specific code in presets (#59) Sina Afrooze 2018-11-06 07:39:29 -08:00
  • 811152126c Export graph to ONNX (#61) Itai Caspi 2018-11-06 10:55:21 +02:00
  • d75df17d97 Modifying ScreenLogger to optionally not output color codes (#56) Leo Dirac 2018-11-05 15:25:49 -08:00
  • 7e7006305a Integrate coach.py params with distributed Coach. (#42) Balaji Subramaniam 2018-11-05 09:33:30 -08:00
  • 95b4fc6888 Added ability to switch between tensorflow and mxnet using -f commandline argument. (#48) Sina Afrooze 2018-10-30 15:29:34 -07:00
  • 2046358ab0 Add docstring for architecture (#47) Sina Afrooze 2018-10-30 02:02:37 -07:00
  • 324c67d614 Bug fix: Removed reference to args which is out of scope. Conditioning now performed one level above. (#54) Thom Lane 2018-10-29 22:29:22 -07:00
  • a888226641 Move embedder, middleware, and head parameters to framework agnostic modules. (#45) Sina Afrooze 2018-10-29 14:46:40 -07:00
  • 16b3e99f37 Setup basic CI flow (#38) Ajay Deshpande 2018-10-24 18:27:58 -07:00
  • 2cc6abc3c4 update CartPole_PPO not addressed during rebase (#41) Zach Dwiel 2018-10-24 19:58:25 -04:00
  • f835ac902c fix renaming: save_checkpoint_sec -> checkpoint_save_secs zach dwiel 2018-10-24 10:52:18 -04:00
  • 78cf25c09a Removing mjkey, should be injected from env var Ajay Deshpande 2018-10-22 14:46:38 -07:00
  • fb2721fffa Removing comments Ajay Deshpande 2018-10-17 17:36:32 -07:00
  • 9a30c26469 Adding improvements Ajay Deshpande 2018-10-15 15:57:10 -07:00
  • 3ba0df7d07 update GraphManager.act specified return type zach dwiel 2018-10-11 20:03:10 -04:00
  • def76b4cc6 update CartPole_PPO zach dwiel 2018-10-11 12:37:09 -04:00
  • 3e5e5475de update training worker zach dwiel 2018-10-11 12:29:57 -04:00
  • 430ca198e5 convert golden tests into pytest format zach dwiel 2018-10-10 16:26:15 -04:00
  • 787ab42578 remove extra call to super().store_episode zach dwiel 2018-10-10 16:20:59 -04:00
  • 7220283653 add len(Episode) Zach Dwiel 2018-10-10 16:17:28 -04:00
  • 700a175902 rename save_checkpoint_secs -> checkpoint_save_secs Zach Dwiel 2018-10-05 11:47:35 -04:00
  • 9804b033a2 rename save_checkpoint_dir -> checkpoint_save_dir Zach Dwiel 2018-10-05 11:44:49 -04:00
  • 201a2237a1 restructure looping mechanism inGraphManager Zach Dwiel 2018-10-05 11:36:42 -04:00
  • 52560a2aae introduce property GraphManager.current_step_counter Zach Dwiel 2018-10-04 17:01:21 -04:00
  • 776c94d551 reorder methods in GraphManager Zach Dwiel 2018-10-04 16:47:41 -04:00
  • 496a516de1 rename GraphManager.sync_graph -> sync Zach Dwiel 2018-10-04 16:43:38 -04:00
  • 5fee48dcfd remove argument keep_networks_in_sync from GraphManager.act, and move this feature into the only place that activated it: GraphManager.train_and_act Zach Dwiel 2018-10-04 11:59:05 -04:00
  • b2d864a5bd remove out of date documentation Zach Dwiel 2018-10-04 11:53:14 -04:00
  • d32d909238 move only invocation of GraphManager.handle_episode_ended inline Zach Dwiel 2018-10-04 11:52:24 -04:00
  • 18d84c5037 remove unnecessary timers from GraphManager Zach Dwiel 2018-10-04 11:49:05 -04:00
  • cd30efe52e remove unnecessary test result is None in GraphManager.act Zach Dwiel 2018-10-04 11:46:51 -04:00
  • 35d67cbd9b use phase context in GraphManager.evaluate Zach Dwiel 2018-10-04 11:38:45 -04:00
  • d3c341147e simplify GraphManager.act by removing arguments: continue_until_game_over and return_on_game_over Zach Dwiel 2018-10-04 11:36:39 -04:00
  • 8be980912c fixed typo from earlier commit Zach Dwiel 2018-10-03 16:53:52 -04:00
  • fbaf19543e capture stdout during preset tests Zach Dwiel 2018-10-03 14:57:57 -04:00
  • 517aac163a introduce graph_manager.phase_context; make sure that calls to graph_manager.train automatically set training phase Zach Dwiel 2018-10-02 16:19:49 -04:00
  • 7382a142bb remove unused steps parameter from GraphManager.train Zach Dwiel 2018-10-02 15:29:57 -04:00
  • 97f608ee5e reorder failing presets Zach Dwiel 2018-10-02 13:28:26 -04:00
  • ad68fa263d remove property GraphManager.training_start_time Zach Dwiel 2018-10-02 13:25:45 -04:00
  • bfc320cf83 disable failing tests for now Zach Dwiel 2018-10-02 13:23:19 -04:00
  • 01f3a0594b remove return values from GraphManager.act Zach Dwiel 2018-10-02 12:44:34 -04:00
  • b02f269464 graph_manager:heatup uses total_steps_counters looping mechanism like other loops. graph_manager:act no longer needs to return any values Zach Dwiel 2018-10-01 17:11:11 -04:00
  • ca9015d8b1 Make NFS work end-to-end. Balaji Subramaniam 2018-10-09 15:49:05 -07:00
  • fb1039fcb5 Checkpoint and evaluation optimizations Ajay Deshpande 2018-10-08 17:49:40 -07:00
  • b285a02023 Adding parameteres, checking transitions before training Ajay Deshpande 2018-10-08 14:49:46 -07:00
  • 0f46877d7e Adding steps and waiting for new checkpoint Ajay Deshpande 2018-10-08 13:41:51 -07:00
  • 0e121c5762 Ignoring redis sub if testing Ajay Deshpande 2018-10-08 12:01:12 -07:00
  • 7f00235ed5 waiting for a new checkpoint if it's available Ajay Deshpande 2018-10-05 19:08:24 -07:00
  • 5eac0102de Changing exception type Ajay Deshpande 2018-10-05 15:57:50 -07:00
  • a7f5442015 Adding should_train helper and should_train in graph_manager Ajay Deshpande 2018-10-05 14:22:15 -07:00
  • a2e57a44f1 Getting only the model_checkpoint_path files Ajay Deshpande 2018-10-05 13:48:10 -07:00
  • 052bbc8f19 Adding lock in s3 Ajay Deshpande 2018-10-05 12:53:51 -07:00
  • 844a5af831 Make distributed coach work end-to-end. - With data store, memory backend and orchestrator interfaces. Balaji Subramaniam 2018-10-04 12:28:21 -07:00