7856d22b09561e33522bdc0bd00218ae75b84bd7,examples/multigoal_sac.py,,run,#Any#,15

Before Change

    }

    M = 128
    qf1 = NNQFunction(
        observation_shape=env.observation_space.shape,
        action_shape=env.action_space.shape,
        hidden_layer_sizes=[M, M],
        name="qf1")
    qf2 = NNQFunction(
        observation_shape=env.observation_space.shape,
        action_shape=env.action_space.shape,
        hidden_layer_sizes=[M, M],

After Change

    M = 128

    q_functions = tuple(
        NNQFunction(
            observation_shape=env.observation_space.shape,
            action_shape=env.action_space.shape,
            hidden_layer_sizes=(M, M),
            name="qf{}".format(i))
        for i in range(2))
    vf = NNVFunction(
        observation_shape=env.observation_space.shape,
        hidden_layer_sizes=[M, M])

    if variant["policy_type"] == "gmm":
        policy = GMMPolicy(
            observation_shape=env.observation_space.shape,
            action_shape=env.action_space.shape,
            K=4,
            hidden_layer_sizes=[M, M],
            qf=q_functions[0],
            reg=0.001
        )
    elif variant["policy_type"] == "lsp":
        bijector_config = {
            "scale_regularization": 0.0,
            "num_coupling_layers": 2,
            "translation_hidden_sizes": (M,),
            "scale_hidden_sizes": (M,),
        }

        policy = LatentSpacePolicy(
            observation_shape=env.observation_space.shape,
            action_shape=env.action_space.shape,
            mode="train",
            squash=True,
            bijector_config=bijector_config,
            observations_preprocessor=None,
            q_function=q_functions[0]
        )

    plotter = QFPolicyPlotter(
        qf=q_functions[0],
        policy=policy,
        obs_lst=np.array([[-2.5, 0.0],
                          [0.0, 0.0],

Instances