7dcda01b219a056b6932096648ad6a53f164811b,samples/dqn_tweaks_doom.py,,,#,62

Before Change

    action_selector = ActionSelectorEpsilonGreedy(epsilon=run.getfloat("defaults", "epsilon"), params=params)
    target_net = agent.TargetNet(model)
    dqn_agent = agent.DQNAgent(dqn_model=target_net.target_model, action_selector=action_selector)
    exp_source = experience.ExperienceSource(env=env_pool, agent=dqn_agent, steps_count=run.getint("defaults", "n_steps"))
    exp_replay = experience.ExperienceReplayBuffer(exp_source, buffer_size=run.getint("exp_buffer", "size"))

    def batch_to_train(batch):

After Change

    action_selector = ActionSelectorEpsilonGreedy(epsilon=run.getfloat("defaults", "epsilon"), params=params)
    target_net = agent.TargetNet(model)
    dqn_agent = agent.DQNAgent(dqn_model=model, action_selector=action_selector)
    exp_source = experience.ExperienceSource(env=env_pool, agent=dqn_agent, steps_count=run.getint("defaults", "n_steps"))
    exp_replay = experience.ExperienceReplayBuffer(exp_source, buffer_size=run.getint("exp_buffer", "size"))

    if run.getboolean("dqn", "target_dqn", fallback=False):

Instances