perf notes about a3c

examples/A3C-Gym/README.md

@@ -20,9 +20,10 @@ because the bottleneck in this implementation is not computation but data.
 
 Some practicical notes:
 
-1. Occasionally, processes may not get terminated completely. It is suggested to use `systemd-run` to run any
+1. Prefer Python 3.
+2. Occasionally, processes may not get terminated completely. It is suggested to use `systemd-run` to run any
 multiprocess Python program to get a cgroup dedicated for the task.
-2. Training with a significant slower speed (e.g. on CPU) will result in very bad score, probably because of async issues.
+3. Training with a significant slower speed (e.g. on CPU) will result in very bad score, probably because of async issues.
 
 ### To run a pretrained Atari model for 100 episodes:
 

tensorpack/predict/concurrency.py

@@ -155,6 +155,7 @@ class MultiThreadAsyncPredictor(AsyncPredictorBase):
             # TODO XXX set logging here to avoid affecting TF logging
             import tornado.options as options
             options.parse_command_line(['--logging=debug'])
+            logger.warn("MultiThreadAsyncPredictor is inefficient in Python 2! Switch to Python 3 instead.")
 
     def start(self):
         if self._need_default_sess: