732197e23a937b7b6d196936519c16ec6317ea9f,rllib/agents/dqn/simple_q.py,,execution_plan,#Any#Any#,116

Before Change

    // (2) Read and train on experiences from the replay buffer.
    replay_op = Replay(local_buffer=local_replay_buffer) \
        .for_each(TrainOneStep(workers)) \
        .for_each(UpdateTargetNetwork(
            workers, config["target_network_update_freq"]))

After Change

    store_op = rollouts.for_each(
        StoreToReplayBuffer(local_buffer=local_replay_buffer))

    if config["simple_optimizer"]:
        train_step_op = TrainOneStep(workers)
    else:
        train_step_op = TrainTFMultiGPU(
            workers=workers,
            sgd_minibatch_size=config["train_batch_size"],
            num_sgd_iter=1,
            num_gpus=config["num_gpus"],
            shuffle_sequences=True,
            _fake_gpus=config["_fake_gpus"],
            framework=config.get("framework"))

    // (2) Read and train on experiences from the replay buffer.
    replay_op = Replay(local_buffer=local_replay_buffer) \
        .for_each(train_step_op) \
        .for_each(UpdateTargetNetwork(
            workers, config["target_network_update_freq"]))

Instances