ML Agents: Use pb file in python with tensorflow to get AgentActions

I use ML-Agents in unity to train a simple linefollow robot. For that i use visual observation with one camera and vector actions with space type discrete (1 branch with 4 options). The training works and i got my nn file as well as the matching pb file. Now i want to use that pb file in a python script to run the same prediction outside of unity. I'm completely new to tensorflow and do not know how i can reproduce my result that i get in float[] vectorAction (values between 0 and 3).

The code I use to start a session:

     with tf.Session() as sess:
         print("load graph")
         with tf.gfile.GFile(GRAPH_PB_PATH, 'rb') as f:
             graph_def = tf.GraphDef()
         graph_def.ParseFromString(f.read())
         sess.graph.as_default()
         tf.import_graph_def(graph_def, name='')
         graph_nodes = [n for n in graph_def.node]
         names = []
         for t in graph_nodes:
             names.append(t.name)
         print(names)
         logits = sess.graph.get_tensor_by_name("visual_observation_0:0")

The code I use for a prediction:

 cv2_edges = numpy.array(cv2_edges).reshape(1, 30, 80, 1)
 prediciton = self.session.run(self.logits, {logits: cv2_edges})

When I print my prediction I get this as a result:

 Prediction: [[[ 64.]
   [  0.]
   [  0.]
   ...
   [  0.]
   [  0.]
   [  0.]]
 
  [[128.]
   [  0.]
   [  0.]
   ...
   [  0.]
   [  0.]
   [  0.]]
 
  [[128.]
   [  0.]
   [  0.]
   ...
   [  0.]
   [  0.]
   [  0.]]
 
  ...
 
  [[  0.]
   [  0.]
   [  0.]
   ...
   [  0.]
   [  0.]
   [  0.]]
 
  [[  0.]
   [  0.]
   [  0.]
   ...
   [  0.]
   [  0.]
   [  0.]]
 
  [[  0.]
   [  0.]
   [  0.]
   ...
   [  0.]
   [  0.]
   [  0.]]]

Maybe it's a stupid question, but i was expection a float[] where I can perform a similar operation like Mathf.FloorToInt() in order to get my discrete action. Do you know how i can get to my expected result?