fix random policy

examples/nop_cartpole.py

@@ -0,0 +1,26 @@
+from rllab.algos.nop import NOP
+from rllab.baselines.zero_baseline import ZeroBaseline
+from rllab.envs.box2d.cartpole_env import CartpoleEnv
+from rllab.envs.normalized_env import normalize
+from rllab.policies.uniform_control_policy import UniformControlPolicy
+
+env = normalize(CartpoleEnv())
+
+policy = UniformControlPolicy(
+    env_spec=env.spec,
+    # The neural network policy should have two hidden layers, each with 32 hidden units.
+)
+
+baseline = ZeroBaseline(env_spec=env.spec)
+
+algo = NOP(
+    env=env,
+    policy=policy,
+    baseline=baseline,
+    batch_size=4000,
+    max_path_length=100,
+    n_itr=40,
+    discount=0.99,
+    step_size=0.01,
+)
+algo.train()

rllab/policies/uniform_control_policy.py

@@ -1,5 +1,6 @@
 from rllab.core.parameterized import Parameterized
 from rllab.core.serializable import Serializable
+from rllab.distributions.delta import Delta
 from rllab.policies.base import Policy
 from rllab.misc.overrides import overrides
 
@@ -19,3 +20,17 @@ def get_action(self, observation):
     def get_params_internal(self, **tags):
         return []
 
+    def get_actions(self, observations):
+        return self.action_space.sample_n(len(observations)), dict()
+
+    @property
+    def vectorized(self):
+        return True
+
+    def reset(self, dones=None):
+        pass
+
+    @property
+    def distribution(self):
+        # Just a placeholder
+        return Delta()