Baselines: The Atari pretrained model downloader be removed?

[louchenyao]

In DQN's README, it gives a way to download the pretrained model:

python -m baselines.deepq.experiments.atari.download_model

But it seems the atari.download_model already be removed in experiments directory.

[unixpickle]

I think this was removed because it was out of date and stopped working. You can browse a different commit to find the script and try it.

Ideally, this should be removed from the README if it indeed no longer works. Otherwise, it should be added back to the repo.

[yrevar]

Is there any alternative? I'm looking for pre-trained model of Pong and few other Atari games for benchmarking and experiments.

[DanielTakeshi]

@yrevar i strongly recommend training them again if you can. They were using outdated atari processing stuff

[louchenyao]

By the way, I can’t find the auto saved model since I trained it by myself. Does anyone have same problem?

[zwfcrazy]

@Chenyao2333 read the code, somehow it's been commented out.

    act = deepq.learn(
        env,
        q_func=model,
        lr=1e-4,
        max_timesteps=args.num_timesteps,
        buffer_size=10000,
        exploration_fraction=0.1,
        exploration_final_eps=0.01,
        train_freq=4,
        learning_starts=10000,
        target_network_update_freq=1000,
        gamma=0.99,
        prioritized_replay=bool(args.prioritized)
    )
    # act.save("pong_model.pkl") XXX
    env.close()