def __init__(self, env): super(ActorCriticNet, self).__init__() self.affine = nn.Linear(env.state_size, 128) self.action_head = nn.Linear(128, env.action_size) self.value_head = nn.Linear(128, 1) self.distribution = distributions.ActionDistribution(env, use_probs=True)
def __init__(self, env, hidden_size=512): super(NatureCNN, self).__init__() self.input_size = 4 self.features = atari.NatureFeatures(self.input_size, hidden_size) self.critic = atari.NatureCritic(hidden_size) self.actor = atari.NatureActor(hidden_size, env.action_size) self.action_dist = distributions.ActionDistribution(env, use_probs=False)
def __init__(self, env): super(ActorCriticNet, self).__init__() self.actor = models.robotics.RoboticsActor(env.state_size, env.action_size, layer_sizes=[64, 64]) self.critic = models.robotics.RoboticsMLP(env.state_size, 1) self.action_dist = dist.ActionDistribution(env, use_probs=False, reparam=False)
def __init__(self, env): super(ActorCriticNet, self).__init__() self.affine1 = nn.Linear(env.state_size['image'], 128) self.action_head = nn.Linear(128, env.action_size) self.value_head = nn.Linear(128, 1) self.distribution = distributions.ActionDistribution(env)