Merge pull request #158 from zuoxingdong/add_ddpg2

update OpenAI-ES

baselines/openaies/experiment.py

@@ -25,17 +25,17 @@
 from lagom.envs.wrappers import VecMonitor
 from lagom.envs.wrappers import VecStandardizeObservation
 
-from openaies import OpenAIES
-from agent import Agent
+from .openaies import OpenAIES
+from .agent import Agent
 
 
 config = Config(
     {'cuda': False, 
      'log.dir': 'logs/default', 
      'log.freq': 10, 
-     'checkpoint.freq': 50,
+     'checkpoint.num': 3,
 
-     'env.id': 'HalfCheetah-v3',#Grid(['HalfCheetah-v3', 'Hopper-v3', 'Walker2d-v3', 'Swimmer-v3']), 
+     'env.id': Grid(['HalfCheetah-v3', 'Hopper-v3', 'Walker2d-v3', 'Swimmer-v3']), 
      'env.standardize_obs': False,
 
      'nn.sizes': [64, 64],
@@ -76,7 +76,7 @@ def initializer(config, seed, device):
     env = make_env(config, seed)
     env = VecMonitor(env)
     if config['env.standardize_obs']:
-        env = VecStandardizeObservation(env, clip=10.)
+        env = VecStandardizeObservation(env, clip=5.)
     global agent
     agent = Agent(config, env, device)
 
@@ -114,12 +114,13 @@ def run(config, seed, device):
                'antithetic': config['train.antithetic'],
                'rank_transform': config['train.rank_transform']})
     train_logs = []
+    checkpoint_count = 0
     with ProcessPoolExecutor(max_workers=config['train.popsize'], initializer=initializer, initargs=(config, seed, device)) as executor:
         print('Finish initialization. Training starts...')
         for generation in range(config['train.generations']):
             start_time = time.perf_counter()
             solutions = es.ask()
-            out = list(executor.map(fitness, solutions))
+            out = list(executor.map(fitness, solutions, chunksize=4))
             Rs, Hs = zip(*out)
             es.tell(solutions, [-R for R in Rs])
             logger = Logger()
@@ -131,9 +132,10 @@ def run(config, seed, device):
             train_logs.append(logger.logs)
             if generation == 0 or (generation+1)%config['log.freq'] == 0:
                 logger.dump(keys=None, index=0, indent=0, border='-'*50)
-            if generation == 0 or (generation+1)%config['checkpoint.freq'] == 0:
+            if (generation+1) >= int(config['train.generations']*(checkpoint_count/(config['checkpoint.num'] - 1))):
                 agent.from_vec(torch.from_numpy(es.result.xbest).float())
                 agent.checkpoint(logdir, generation+1)
+                checkpoint_count += 1
     pickle_dump(obj=train_logs, f=logdir/'train_logs', ext='.pkl')
     return None
 
@@ -142,4 +144,4 @@ def run(config, seed, device):
     run_experiment(run=run, 
                    config=config, 
                    seeds=[1770966829, 1500925526, 2054191100], 
-                   num_worker=os.cpu_count())
+                   num_worker=5)

baselines/openaies/logs/default/0/1500925526/anim/openaigym.episode_batch.0.1855909.stats.json

@@ -0,0 +1 @@
+{"initial_reset_timestamp": 1557149739.7226458, "timestamps": [1557149748.523486], "episode_lengths": [1000], "episode_rewards": [1912.6200380948023], "episode_types": ["t", "t"]}

baselines/openaies/logs/default/0/1500925526/anim/openaigym.manifest.0.1855909.manifest.json

@@ -0,0 +1 @@
+{"stats": "openaigym.episode_batch.0.1855909.stats.json", "videos": [["openaigym.video.0.1855909.video000000.mp4", "openaigym.video.0.1855909.video000000.meta.json"], ["openaigym.video.0.1855909.video000001.mp4", "openaigym.video.0.1855909.video000001.meta.json"]], "env_info": {"gym_version": "0.12.1", "env_id": "HalfCheetah-v3"}}

baselines/openaies/logs/default/0/1500925526/anim/openaigym.video.0.1855909.video000000.meta.json

@@ -0,0 +1 @@
+{"episode_id": 0, "content_type": "video/mp4", "encoder_version": {"backend": "ffmpeg", "version": "b'ffmpeg version 3.4.4-0ubuntu0.18.04.1 Copyright (c) 2000-2018 the FFmpeg developers\\nbuilt with gcc 7 (Ubuntu 7.3.0-16ubuntu3)\\nconfiguration: --prefix=/usr --extra-version=0ubuntu0.18.04.1 --toolchain=hardened --libdir=/usr/lib/x86_64-linux-gnu --incdir=/usr/include/x86_64-linux-gnu --enable-gpl --disable-stripping --enable-avresample --enable-avisynth --enable-gnutls --enable-ladspa --enable-libass --enable-libbluray --enable-libbs2b --enable-libcaca --enable-libcdio --enable-libflite --enable-libfontconfig --enable-libfreetype --enable-libfribidi --enable-libgme --enable-libgsm --enable-libmp3lame --enable-libmysofa --enable-libopenjpeg --enable-libopenmpt --enable-libopus --enable-libpulse --enable-librubberband --enable-librsvg --enable-libshine --enable-libsnappy --enable-libsoxr --enable-libspeex --enable-libssh --enable-libtheora --enable-libtwolame --enable-libvorbis --enable-libvpx --enable-libwavpack --enable-libwebp --enable-libx265 --enable-libxml2 --enable-libxvid --enable-libzmq --enable-libzvbi --enable-omx --enable-openal --enable-opengl --enable-sdl2 --enable-libdc1394 --enable-libdrm --enable-libiec61883 --enable-chromaprint --enable-frei0r --enable-libopencv --enable-libx264 --enable-shared\\nlibavutil      55. 78.100 / 55. 78.100\\nlibavcodec     57.107.100 / 57.107.100\\nlibavformat    57. 83.100 / 57. 83.100\\nlibavdevice    57. 10.100 / 57. 10.100\\nlibavfilter     6.107.100 /  6.107.100\\nlibavresample   3.  7.  0 /  3.  7.  0\\nlibswscale      4.  8.100 /  4.  8.100\\nlibswresample   2.  9.100 /  2.  9.100\\nlibpostproc    54.  7.100 / 54.  7.100\\n'", "cmdline": ["ffmpeg", "-nostats", "-loglevel", "error", "-y", "-r", "20", "-f", "rawvideo", "-s:v", "500x500", "-pix_fmt", "rgb24", "-i", "-", "-vf", "scale=trunc(iw/2)*2:trunc(ih/2)*2", "-vcodec", "libx264", "-pix_fmt", "yuv420p", "/home/zuo/Code/tmp/lagom/baselines/openaies/logs/default/0/1500925526/anim/openaigym.video.0.1855909.video000000.mp4"]}}

baselines/openaies/logs/default/0/1500925526/anim/openaigym.video.0.1855909.video000001.meta.json

@@ -0,0 +1 @@
+{"episode_id": 1, "content_type": "video/mp4", "encoder_version": {"backend": "ffmpeg", "version": "b'ffmpeg version 3.4.4-0ubuntu0.18.04.1 Copyright (c) 2000-2018 the FFmpeg developers\\nbuilt with gcc 7 (Ubuntu 7.3.0-16ubuntu3)\\nconfiguration: --prefix=/usr --extra-version=0ubuntu0.18.04.1 --toolchain=hardened --libdir=/usr/lib/x86_64-linux-gnu --incdir=/usr/include/x86_64-linux-gnu --enable-gpl --disable-stripping --enable-avresample --enable-avisynth --enable-gnutls --enable-ladspa --enable-libass --enable-libbluray --enable-libbs2b --enable-libcaca --enable-libcdio --enable-libflite --enable-libfontconfig --enable-libfreetype --enable-libfribidi --enable-libgme --enable-libgsm --enable-libmp3lame --enable-libmysofa --enable-libopenjpeg --enable-libopenmpt --enable-libopus --enable-libpulse --enable-librubberband --enable-librsvg --enable-libshine --enable-libsnappy --enable-libsoxr --enable-libspeex --enable-libssh --enable-libtheora --enable-libtwolame --enable-libvorbis --enable-libvpx --enable-libwavpack --enable-libwebp --enable-libx265 --enable-libxml2 --enable-libxvid --enable-libzmq --enable-libzvbi --enable-omx --enable-openal --enable-opengl --enable-sdl2 --enable-libdc1394 --enable-libdrm --enable-libiec61883 --enable-chromaprint --enable-frei0r --enable-libopencv --enable-libx264 --enable-shared\\nlibavutil      55. 78.100 / 55. 78.100\\nlibavcodec     57.107.100 / 57.107.100\\nlibavformat    57. 83.100 / 57. 83.100\\nlibavdevice    57. 10.100 / 57. 10.100\\nlibavfilter     6.107.100 /  6.107.100\\nlibavresample   3.  7.  0 /  3.  7.  0\\nlibswscale      4.  8.100 /  4.  8.100\\nlibswresample   2.  9.100 /  2.  9.100\\nlibpostproc    54.  7.100 / 54.  7.100\\n'", "cmdline": ["ffmpeg", "-nostats", "-loglevel", "error", "-y", "-r", "20", "-f", "rawvideo", "-s:v", "500x500", "-pix_fmt", "rgb24", "-i", "-", "-vf", "scale=trunc(iw/2)*2:trunc(ih/2)*2", "-vcodec", "libx264", "-pix_fmt", "yuv420p", "/home/zuo/Code/tmp/lagom/baselines/openaies/logs/default/0/1500925526/anim/openaigym.video.0.1855909.video000001.mp4"]}}

baselines/openaies/logs/default/0/config.yml

@@ -2,8 +2,7 @@ ID: 0
 cuda: false
 log.dir: logs/default
 log.freq: 10
-checkpoint.freq: 50
-env.id: HalfCheetah-v3
+checkpoint.num: 3
 env.standardize_obs: false
 nn.sizes:
 - 64
@@ -24,3 +23,4 @@ train.sigma_scheduler_args:
 - 0
 train.antithetic: false
 train.rank_transform: true
+env.id: HalfCheetah-v3

baselines/openaies/logs/default/1/config.yml

@@ -0,0 +1,26 @@
+ID: 1
+cuda: false
+log.dir: logs/default
+log.freq: 10
+checkpoint.num: 3
+env.standardize_obs: false
+nn.sizes:
+- 64
+- 64
+env.clip_action: true
+agent.std0: 0.5
+train.generations: 1000
+train.popsize: 64
+train.mu0: 0.0
+train.std0: 1.0
+train.lr: 0.1
+train.lr_decay: 1.0
+train.min_lr: 1.0e-06
+train.sigma_scheduler_args:
+- 1.0
+- 0.01
+- 450
+- 0
+train.antithetic: false
+train.rank_transform: true
+env.id: Hopper-v3

baselines/openaies/logs/default/2/config.yml

@@ -0,0 +1,26 @@
+ID: 2
+cuda: false
+log.dir: logs/default
+log.freq: 10
+checkpoint.num: 3
+env.standardize_obs: false
+nn.sizes:
+- 64
+- 64
+env.clip_action: true
+agent.std0: 0.5
+train.generations: 1000
+train.popsize: 64
+train.mu0: 0.0
+train.std0: 1.0
+train.lr: 0.1
+train.lr_decay: 1.0
+train.min_lr: 1.0e-06
+train.sigma_scheduler_args:
+- 1.0
+- 0.01
+- 450
+- 0
+train.antithetic: false
+train.rank_transform: true
+env.id: Walker2d-v3

baselines/openaies/logs/default/3/config.yml

@@ -0,0 +1,26 @@
+ID: 3
+cuda: false
+log.dir: logs/default
+log.freq: 10
+checkpoint.num: 3
+env.standardize_obs: false
+nn.sizes:
+- 64
+- 64
+env.clip_action: true
+agent.std0: 0.5
+train.generations: 1000
+train.popsize: 64
+train.mu0: 0.0
+train.std0: 1.0
+train.lr: 0.1
+train.lr_decay: 1.0
+train.min_lr: 1.0e-06
+train.sigma_scheduler_args:
+- 1.0
+- 0.01
+- 450
+- 0
+train.antithetic: false
+train.rank_transform: true
+env.id: Swimmer-v3

baselines/openaies/logs/default/source_files/experiment.py

@@ -33,9 +33,9 @@
     {'cuda': False, 
      'log.dir': 'logs/default', 
      'log.freq': 10, 
-     'checkpoint.freq': 50,
+     'checkpoint.num': 3,
 
-     'env.id': 'HalfCheetah-v3',#Grid(['HalfCheetah-v3', 'Hopper-v3', 'Walker2d-v3', 'Swimmer-v3']), 
+     'env.id': Grid(['HalfCheetah-v3', 'Hopper-v3', 'Walker2d-v3', 'Swimmer-v3']), 
      'env.standardize_obs': False,
 
      'nn.sizes': [64, 64],
@@ -76,7 +76,7 @@ def initializer(config, seed, device):
     env = make_env(config, seed)
     env = VecMonitor(env)
     if config['env.standardize_obs']:
-        env = VecStandardizeObservation(env, clip=10.)
+        env = VecStandardizeObservation(env, clip=5.)
     global agent
     agent = Agent(config, env, device)
 
@@ -114,12 +114,13 @@ def run(config, seed, device):
                'antithetic': config['train.antithetic'],
                'rank_transform': config['train.rank_transform']})
     train_logs = []
+    checkpoint_count = 0
     with ProcessPoolExecutor(max_workers=config['train.popsize'], initializer=initializer, initargs=(config, seed, device)) as executor:
         print('Finish initialization. Training starts...')
         for generation in range(config['train.generations']):
             start_time = time.perf_counter()
             solutions = es.ask()
-            out = list(executor.map(fitness, solutions))
+            out = list(executor.map(fitness, solutions, chunksize=4))
             Rs, Hs = zip(*out)
             es.tell(solutions, [-R for R in Rs])
             logger = Logger()
@@ -131,9 +132,10 @@ def run(config, seed, device):
             train_logs.append(logger.logs)
             if generation == 0 or (generation+1)%config['log.freq'] == 0:
                 logger.dump(keys=None, index=0, indent=0, border='-'*50)
-            if generation == 0 or (generation+1)%config['checkpoint.freq'] == 0:
+            if (generation+1) >= int(config['train.generations']*(checkpoint_count/(config['checkpoint.num'] - 1))):
                 agent.from_vec(torch.from_numpy(es.result.xbest).float())
                 agent.checkpoint(logdir, generation+1)
+                checkpoint_count += 1
     pickle_dump(obj=train_logs, f=logdir/'train_logs', ext='.pkl')
     return None
 
@@ -142,4 +144,4 @@ def run(config, seed, device):
     run_experiment(run=run, 
                    config=config, 
                    seeds=[1770966829, 1500925526, 2054191100], 
-                   num_worker=os.cpu_count())
+                   num_worker=5)