put filters and running_stat files in common instead of acktr (#389)

baselines/common/runners.py

@@ -0,0 +1,18 @@
+import numpy as np
+from abc import ABC, abstractmethod
+
+class AbstractEnvRunner(ABC):
+    def __init__(self, *, env, model, nsteps):
+        self.env = env
+        self.model = model
+        nenv = env.num_envs
+        self.batch_ob_shape = (nenv*nsteps,) + env.observation_space.shape
+        self.obs = np.zeros((nenv,) + env.observation_space.shape, dtype=model.train_model.X.dtype.name)
+        self.obs[:] = env.reset()
+        self.nsteps = nsteps
+        self.states = model.initial_state
+        self.dones = [False for _ in range(nenv)]
+
+    @abstractmethod
+    def run(self):
+        raise NotImplementedError