MCPcopy Create free account
hub / github.com/pytorch/examples / run_episode

Method run_episode

distributed/rpc/rl/main.py:90–110  ·  view source on GitHub ↗

r""" Run one episode of n_steps. Args: agent_rref (RRef): an RRef referencing the agent object. n_steps (int): number of steps in this episode

(self, agent_rref, n_steps)

Source from the content-addressed store, hash-verified

88 self.env.reset(seed=args.seed)
89
90 def run_episode(self, agent_rref, n_steps):
91 r"""
92 Run one episode of n_steps.
93
94 Args:
95 agent_rref (RRef): an RRef referencing the agent object.
96 n_steps (int): number of steps in this episode
97 """
98 state, ep_reward = self.env.reset()[0], 0
99 for step in range(n_steps):
100 # send the state to the agent to get an action
101 action = _remote_method(Agent.select_action, agent_rref, self.id, state)
102
103 # apply the action to the environment, and get the reward
104 state, reward, terminated, truncated, _ = self.env.step(action)
105
106 # report the reward to the agent for training purpose
107 _remote_method(Agent.report_reward, agent_rref, self.id, reward)
108
109 if terminated or truncated:
110 break
111
112class Agent:
113 def __init__(self, world_size):

Callers

nothing calls this directly

Calls 2

resetMethod · 0.80
_remote_methodFunction · 0.70

Tested by

no test coverage detected