Adding should_train helper and should_train in graph_manager

rl_coach/agents/agent.py

@@ -523,6 +523,8 @@ class Agent(AgentInterface):
         Determine if online weights should be copied to the target.
         :return: boolean: True if the online weights should be copied to the target.
         """
+        if hasattr(self.ap.memory, 'memory_backend_params'):
+            self.total_steps_counter = self.call_memory('num_transitions')
         # update the target network of every network that has a target network
         step_method = self.ap.algorithm.num_steps_between_copying_online_weights_to_target
         if step_method.__class__ == TrainingSteps:
@@ -544,22 +546,35 @@ class Agent(AgentInterface):
         :return:  boolean: True if we should start a training phase
         """
 
+        should_update = self._should_train_helper(wait_for_full_episode)
+
+        step_method = self.ap.algorithm.num_consecutive_playing_steps
+
+        if should_update:
+            if step_method.__class__ == EnvironmentEpisodes:
+                self.last_training_phase_step = self.current_episode
+            if step_method.__class__ == EnvironmentSteps:
+                self.last_training_phase_step = self.total_steps_counter
+
+        return should_update
+
+    def _should_train_helper(self, wait_for_full_episode=False):
+
         if hasattr(self.ap.memory, 'memory_backend_params'):
             self.total_steps_counter = self.call_memory('num_transitions')
+
         step_method = self.ap.algorithm.num_consecutive_playing_steps
         if step_method.__class__ == EnvironmentEpisodes:
             should_update = (self.current_episode - self.last_training_phase_step) >= step_method.num_steps
-            if should_update:
+
-                self.last_training_phase_step = self.current_episode
         elif step_method.__class__ == EnvironmentSteps:
             should_update = (self.total_steps_counter - self.last_training_phase_step) >= step_method.num_steps
             if wait_for_full_episode:
                 should_update = should_update and self.current_episode_buffer.is_complete
-            if should_update:
-                self.last_training_phase_step = self.total_steps_counter
         else:
             raise ValueError("The num_consecutive_playing_steps parameter should be either "
                              "EnvironmentSteps or Episodes. Instead it is {}".format(step_method.__class__))
+
         return should_update
 
     def train(self):

rl_coach/data_stores/s3_data_store.py

@@ -5,6 +5,7 @@ from minio.error import ResponseError
 from configparser import ConfigParser, Error
 from google.protobuf import text_format
 from tensorflow.python.training.checkpoint_state_pb2 import CheckpointState
+from minio.error import ResponseError
 import os
 import time
 import io
@@ -63,7 +64,7 @@ class S3DataStore(DataStore):
                 for filename in files:
                     if filename == 'checkpoint':
                         checkpoint_file = (root, filename)
-                        pass
+                        continue
                     abs_name = os.path.abspath(os.path.join(root, filename))
                     rel_name = os.path.relpath(abs_name, self.params.checkpoint_dir)
                     self.mc.fput_object(self.params.bucket_name, rel_name, abs_name)
@@ -79,17 +80,21 @@ class S3DataStore(DataStore):
     def load_from_store(self):
         try:
 
+            filename = os.path.abspath(os.path.join(self.params.checkpoint_dir, "checkpoint"))
+
             while True:
                 objects = self.mc.list_objects_v2(self.params.bucket_name, self.params.lock_file)
-                time.sleep(10)
+
                 if next(objects, None) is None:
+                    try:
+                        self.mc.fget_object(self.params.bucket_name, "checkpoint", filename)
+                    except ResponseError as e:
+                        continue
                     break
+                time.sleep(10)
 
             print("loading from s3")
 
-            filename = os.path.abspath(os.path.join(self.params.checkpoint_dir, "checkpoint"))
-            self.mc.fget_object(self.params.bucket_name, "checkpoint", filename)
-
             ckpt = CheckpointState()
             if os.path.exists(filename):
                 contents = open(filename, 'r').read()

rl_coach/graph_managers/graph_manager.py

@@ -591,3 +591,6 @@ class GraphManager(object):
             result += "{}: \n{}\n".format(key, params)
 
         return result
+
+    def should_train(self) -> bool:
+        return any([manager.should_train() for manager in self.level_managers])

rl_coach/level_manager.py

@@ -260,3 +260,6 @@ class LevelManager(EnvironmentInterface):
         :return:
         """
         [agent.sync() for agent in self.agents.values()]
+
+    def should_train(self) -> bool:
+        return any([agent._should_train_helper() for agent in self.agents.values()])

rl_coach/orchestrators/start_training.py

@@ -20,12 +20,12 @@ def main(preset: str, image: str='ajaysudh/testing:coach', num_workers: int=1, n
     if data_store == "s3":
         ds_params = DataStoreParameters("s3", "", "")
         ds_params_instance = S3DataStoreParameters(ds_params=ds_params, end_point=s3_end_point, bucket_name=s3_bucket_name,
-                                             checkpoint_dir="/checkpoint")
+                                                   checkpoint_dir="/checkpoint")
     elif data_store == "nfs":
         ds_params = DataStoreParameters("nfs", "kubernetes", {"namespace": "default"})
         ds_params_instance = NFSDataStoreParameters(ds_params)
 
-    worker_run_type_params = RunTypeParameters(image, rollout_command, run_type="worker")
+    worker_run_type_params = RunTypeParameters(image, rollout_command, run_type="worker", num_replicas=num_workers)
     trainer_run_type_params = RunTypeParameters(image, training_command, run_type="trainer")
 
     orchestration_params = KubernetesParameters([worker_run_type_params, trainer_run_type_params],
@@ -53,7 +53,7 @@ def main(preset: str, image: str='ajaysudh/testing:coach', num_workers: int=1, n
         orchestrator.trainer_logs()
     except KeyboardInterrupt:
         pass
-    # orchestrator.undeploy()
+    orchestrator.undeploy()
 
 
 if __name__ == '__main__':
@@ -90,6 +90,11 @@ if __name__ == '__main__':
                         help="(string) S3 bucket name to use when S3 data store is used.",
                         type=str,
                         required=True)
+    parser.add_argument('--num-workers',
+                        help="(string) Number of rollout workers",
+                        type=int,
+                        required=False,
+                        default=1)
 
     # parser.add_argument('--checkpoint_dir',
     #                     help='(string) Path to a folder containing a checkpoint to write the model to.',
@@ -99,4 +104,4 @@ if __name__ == '__main__':
 
     main(preset=args.preset, image=args.image, nfs_server=args.nfs_server, nfs_path=args.nfs_path,
          memory_backend=args.memory_backend, data_store=args.data_store, s3_end_point=args.s3_end_point,
-         s3_bucket_name=args.s3_bucket_name)
+         s3_bucket_name=args.s3_bucket_name, num_workers=args.num_workers)

rl_coach/presets/CartPole_PPO.py

@@ -0,0 +1,75 @@
+from rl_coach.agents.clipped_ppo_agent import ClippedPPOAgentParameters
+from rl_coach.architectures.tensorflow_components.architecture import Dense
+from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
+from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, RunPhase
+from rl_coach.environments.environment import MaxDumpMethod, SelectedPhaseOnlyDumpMethod, SingleLevelSelection
+from rl_coach.environments.gym_environment import Mujoco, mujoco_v2, MujocoInputFilter
+from rl_coach.exploration_policies.additive_noise import AdditiveNoiseParameters
+from rl_coach.exploration_policies.e_greedy import EGreedyParameters
+from rl_coach.filters.observation.observation_normalization_filter import ObservationNormalizationFilter
+from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
+from rl_coach.graph_managers.graph_manager import ScheduleParameters
+from rl_coach.schedules import LinearSchedule
+
+####################
+# Graph Scheduling #
+####################
+
+schedule_params = ScheduleParameters()
+schedule_params.improve_steps = TrainingSteps(10000000)
+schedule_params.steps_between_evaluation_periods = EnvironmentSteps(2048)
+schedule_params.evaluation_steps = EnvironmentEpisodes(5)
+schedule_params.heatup_steps = EnvironmentSteps(0)
+
+#########
+# Agent #
+#########
+agent_params = ClippedPPOAgentParameters()
+
+
+agent_params.network_wrappers['main'].learning_rate = 0.0003
+agent_params.network_wrappers['main'].input_embedders_parameters['observation'].activation_function = 'tanh'
+agent_params.network_wrappers['main'].input_embedders_parameters['observation'].scheme = [Dense([64])]
+agent_params.network_wrappers['main'].middleware_parameters.scheme = [Dense([64])]
+agent_params.network_wrappers['main'].middleware_parameters.activation_function = 'tanh'
+agent_params.network_wrappers['main'].batch_size = 64
+agent_params.network_wrappers['main'].optimizer_epsilon = 1e-5
+agent_params.network_wrappers['main'].adam_optimizer_beta2 = 0.999
+
+agent_params.algorithm.clip_likelihood_ratio_using_epsilon = 0.2
+agent_params.algorithm.clipping_decay_schedule = LinearSchedule(1.0, 0, 1000000)
+agent_params.algorithm.beta_entropy = 0
+agent_params.algorithm.gae_lambda = 0.95
+agent_params.algorithm.discount = 0.99
+agent_params.algorithm.optimization_epochs = 10
+agent_params.algorithm.estimate_state_value_using_gae = True
+agent_params.algorithm.num_steps_between_copying_online_weights_to_target = EnvironmentSteps(2048)
+
+# agent_params.input_filter = MujocoInputFilter()
+agent_params.exploration = EGreedyParameters()
+agent_params.exploration.epsilon_schedule = LinearSchedule(1.0, 0.01, 10000)
+# agent_params.pre_network_filter = MujocoInputFilter()
+agent_params.pre_network_filter.add_observation_filter('observation', 'normalize_observation',
+                                                       ObservationNormalizationFilter(name='normalize_observation'))
+
+###############
+# Environment #
+###############
+env_params = Mujoco()
+env_params.level = 'CartPole-v0'
+
+vis_params = VisualizationParameters()
+vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
+vis_params.dump_mp4 = False
+
+########
+# Test #
+########
+preset_validation_params = PresetValidationParameters()
+preset_validation_params.test = True
+preset_validation_params.min_reward_threshold = 150
+preset_validation_params.max_episodes_to_achieve_reward = 250
+
+graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
+                                    schedule_params=schedule_params, vis_params=vis_params,
+                                    preset_validation_params=preset_validation_params)

rl_coach/training_worker.py

@@ -32,13 +32,13 @@ def training_worker(graph_manager, checkpoint_dir):
 
     # training loop
     while True:
-        graph_manager.phase = core_types.RunPhase.TRAIN
+        if graph_manager.should_train():
-        graph_manager.train(core_types.TrainingSteps(1))
+            graph_manager.phase = core_types.RunPhase.TRAIN
-        graph_manager.phase = core_types.RunPhase.UNDEFINED
+            graph_manager.train(core_types.TrainingSteps(1))
-
+            graph_manager.phase = core_types.RunPhase.UNDEFINED
-        graph_manager.evaluate(graph_manager.evaluation_steps)
+            graph_manager.evaluate(graph_manager.evaluation_steps)
-
+            graph_manager.save_checkpoint()
-        graph_manager.save_checkpoint()
+        time.sleep(10)
 
 
 def main():