Cleanup imports.

Till now, most of the modules were importing all of the module objects
(variables, classes, functions, other imports) into module namespace,
which potentially could (and was) cause of unintentional use of class or
methods, which was indirect imported.

With this patch, all the star imports were substituted with top-level
module, which provides desired class or function.

Besides, all imports where sorted (where possible) in a way pep8[1]
suggests - first are imports from standard library, than goes third
party imports (like numpy, tensorflow etc) and finally coach modules.
All of those sections are separated by one empty line.

[1] https://www.python.org/dev/peps/pep-0008/#imports

agents/naf_agent.py

@@ -13,21 +13,20 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
-
 import numpy as np
 
 from agents.value_optimization_agent import ValueOptimizationAgent
-from utils import RunPhase, Signal
+import utils
 
 
 # Normalized Advantage Functions - https://arxiv.org/pdf/1603.00748.pdf
 class NAFAgent(ValueOptimizationAgent):
     def __init__(self, env, tuning_parameters, replicated_device=None, thread_id=0):
         ValueOptimizationAgent.__init__(self, env, tuning_parameters, replicated_device, thread_id)
-        self.l_values = Signal("L")
-        self.a_values = Signal("Advantage")
-        self.mu_values = Signal("Action")
-        self.v_values = Signal("V")
+        self.l_values = utils.Signal("L")
+        self.a_values = utils.Signal("Advantage")
+        self.mu_values = utils.Signal("Action")
+        self.v_values = utils.Signal("V")
         self.signals += [self.l_values, self.a_values, self.mu_values, self.v_values]
 
     def learn_from_batch(self, batch):
@@ -49,7 +48,7 @@ class NAFAgent(ValueOptimizationAgent):
 
         return total_loss
 
-    def choose_action(self, curr_state, phase=RunPhase.TRAIN):
+    def choose_action(self, curr_state, phase=utils.RunPhase.TRAIN):
         assert not self.env.discrete_controls, 'NAF works only for continuous control problems'
 
         # convert to batch so we can run it through the network
@@ -60,7 +59,7 @@ class NAFAgent(ValueOptimizationAgent):
             outputs=naf_head.mu,
             squeeze_output=False,
         )
-        if phase == RunPhase.TRAIN:
+        if phase == utils.RunPhase.TRAIN:
             action = self.exploration_policy.get_action(action_values)
         else:
             action = action_values