f5e8cbc21c5a78e2c2996e256f0f2bd5b6930f90,src/garage/tf/algos/reps.py,REPS,__init__,#REPS#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#,56

Before Change

        self._l2_reg_dual = float(l2_reg_dual)
        self._l2_reg_loss = float(l2_reg_loss)

        super(REPS, self).__init__(env_spec=env_spec,
                                   policy=policy,
                                   baseline=baseline,
                                   max_path_length=max_path_length,
                                   discount=discount,
                                   gae_lambda=gae_lambda,
                                   center_adv=center_adv,
                                   positive_adv=positive_adv,
                                   fixed_horizon=fixed_horizon)

    def init_opt(self):
        Initialize the optimization procedure.
        pol_loss_inputs, pol_opt_inputs, dual_opt_inputs = self._build_inputs()

After Change

        optimizer_args = optimizer_args or dict(max_opt_itr=50)
        dual_optimizer_args = dual_optimizer_args or dict(maxiter=50)

        self.policy = policy
        self.max_path_length = max_path_length

        self._env_spec = env_spec
        self._baseline = baseline
        self._discount = discount
        self._gae_lambda = gae_lambda
        self._center_adv = center_adv
        self._positive_adv = positive_adv
        self._fixed_horizon = fixed_horizon
        self._flatten_input = True

        self._name = name
        self._name_scope = tf.name_scope(self._name)
        self._old_policy = policy.clone("old_policy")

        self._feat_diff = None
        self._param_eta = None
        self._param_v = None
        self._f_dual = None
        self._f_dual_grad = None
        self._f_policy_kl = None

        self._optimizer = optimizer(**optimizer_args)
        self._dual_optimizer = dual_optimizer
        self._dual_optimizer_args = dual_optimizer_args
        self._epsilon = float(epsilon)
        self._l2_reg_dual = float(l2_reg_dual)
        self._l2_reg_loss = float(l2_reg_loss)

        self._episode_reward_mean = collections.deque(maxlen=100)
        if policy.vectorized:
            self.sampler_cls = OnPolicyVectorizedSampler
        else:
            self.sampler_cls = BatchSampler

        self.init_opt()

    def init_opt(self):
        Initialize the optimization procedure.
        pol_loss_inputs, pol_opt_inputs, dual_opt_inputs = self._build_inputs()

Instances