def mlp_actor_critic()

in spinup/algos/tf1/sac/core.py [0:0]

• 14 lines of code
• 1 McCabe index (conditional complexity)

def mlp_actor_critic(x, a, hidden_sizes=(256,256), activation=tf.nn.relu, 
                     output_activation=None, policy=mlp_gaussian_policy, action_space=None):
    # policy
    with tf.variable_scope('pi'):
        mu, pi, logp_pi = policy(x, a, hidden_sizes, activation, output_activation)
        mu, pi, logp_pi = apply_squashing_func(mu, pi, logp_pi)

    # make sure actions are in correct range
    action_scale = action_space.high[0]
    mu *= action_scale
    pi *= action_scale

    # vfs
    vf_mlp = lambda x : tf.squeeze(mlp(x, list(hidden_sizes)+[1], activation, None), axis=1)
    with tf.variable_scope('q1'):
        q1 = vf_mlp(tf.concat([x,a], axis=-1))
    with tf.variable_scope('q2'):
        q2 = vf_mlp(tf.concat([x,a], axis=-1))
    return mu, pi, logp_pi, q1, q2