Adding kubernetes orchestrator for rollouts, adding requirements for incremental docker builds

rl_coach/agents/dqn_agent.py

@@ -27,6 +27,7 @@ from rl_coach.architectures.tensorflow_components.embedders.embedder import Inpu
 from rl_coach.core_types import EnvironmentSteps
 from rl_coach.exploration_policies.e_greedy import EGreedyParameters
 from rl_coach.memories.non_episodic.experience_replay import ExperienceReplayParameters
+from rl_coach.memories.non_episodic.distributed_experience_replay import DistributedExperienceReplayParameters
 from rl_coach.schedules import LinearSchedule
 
 
@@ -50,6 +51,20 @@ class DQNNetworkParameters(NetworkParameters):
         self.create_target_network = True
 
 
+class DQNAgentParametersDistributed(AgentParameters):
+    def __init__(self):
+        super().__init__(algorithm=DQNAlgorithmParameters(),
+                         exploration=EGreedyParameters(),
+                         memory=DistributedExperienceReplayParameters(),
+                         networks={"main": DQNNetworkParameters()})
+        self.exploration.epsilon_schedule = LinearSchedule(1, 0.1, 1000000)
+        self.exploration.evaluation_epsilon = 0.05
+
+    @property
+    def path(self):
+        return 'rl_coach.agents.dqn_agent:DQNAgent'
+
+
 class DQNAgentParameters(AgentParameters):
     def __init__(self):
         super().__init__(algorithm=DQNAlgorithmParameters(),