3a3bdf17ca08052309016bdfa481a58c70112abd,tests/garage/tf/policies/test_categorical_mlp_policy_with_model.py,TestCategoricalMLPPolicyWithModel,test_is_pickleable,#TestCategoricalMLPPolicyWithModel#Any#Any#Any#,101

Before Change

        with tf.Session(graph=tf.Graph()):
            policy_pickled = pickle.loads(p)
            action, prob = policy_pickled.get_action(obs)
            assert env.action_space.contains(action)
            assert action == 0
            assert np.array_equal(prob["prob"], expected_prob)

After Change

        env.reset()
        obs, _, _, _ = env.step(1)

        with tf.variable_scope("CategoricalMLPPolicy/MLPModel", reuse=True):
            return_var = tf.get_variable("return_var")
        // assign it to all one
        return_var.load(tf.ones_like(return_var).eval())
        output1 = self.sess.run(
            policy.model.outputs,
            feed_dict={policy.model.input: [obs.flatten()]})

Instances