f2c3c23b3c174ffdbd9ab24aff9b68ffb2acd94a,REINFORCE.py,,main,#,248

Before Change

        self.session.run(init)

def main():
    if(len(sys.argv) < 3):
        print("Please provide the name of an environment and a path to save monitor files")
        return
    env = gym.make(sys.argv[1])
    rank = len(env.observation_space.shape)  // Observation space rank
    if isinstance(env.action_space, Discrete):
        action_selection = ProbabilisticCategoricalActionSelection()
        if rank == 1:
            agent = REINFORCELearnerDiscrete(env, action_selection, sys.argv[2])
        else:
            agent = REINFORCELearnerDiscreteCNN(env, action_selection, sys.argv[2])
    elif isinstance(env.action_space, Box):
        action_selection = ContinuousActionSelection()
        if rank == 1:
            agent = REINFORCELearnerContinuous(env, action_selection, sys.argv[2])
        else:
            raise NotImplementedError
    else:

After Change

parser.add_argument("--rnn", action="store_true", default=False, help="Use a Recurrent Neural Network (only for envs with a state space of rank 1).")

def main():
    try:
        args = parser.parse_args()
    except:
        sys.exit()
    env = gym.make(args.environment)
    rank = len(env.observation_space.shape)  // Observation space rank
    if isinstance(env.action_space, Discrete):
        action_selection = ProbabilisticCategoricalActionSelection()
        if rank == 1:
            agent = REINFORCELearnerDiscrete(env, action_selection, args.rnn, args.monitor_path)
        else:
            agent = REINFORCELearnerDiscreteCNN(env, action_selection, args.monitor_path)
    elif isinstance(env.action_space, Box):
        action_selection = ContinuousActionSelection()
        if rank == 1:
            agent = REINFORCELearnerContinuous(env, action_selection, args.rnn, args.monitor_path)
        else:
            raise NotImplementedError
    else:
        raise NotImplementedError
    try:
        agent.env = wrappers.Monitor(agent.env, args.monitor_path, force=True)
        agent.learn()
    except KeyboardInterrupt:
        pass

Instances