Error starting Mujoco task with Trust-PCL

[ViktorM]

When trying to start Mujoco task with Trust-PCL as recommended in https://github.com/tensorflow/models/tree/master/research/pcl_rl readme I've hit an error:

[2017-12-03 05:36:46,648] Making new env: Walker2d-v1
Traceback (most recent call last):
  File "trainer.py", line 453, in <module>
    app.run()
  File "/home/viktor/anaconda3/envs/mujoco/lib/python3.5/site-packages/tensorflow/python/platform/app.py", line 48, in run
    _sys.exit(main(_sys.argv[:1] + flags_passthrough))
  File "trainer.py", line 449, in main
    trainer.run()
  File "trainer.py", line 388, in run
    self.controller.setup()
  File "/home/viktor/DeepRL/models/research/pcl_rl/controller.py", line 118, in setup
    self.model.setup()
  File "/home/viktor/DeepRL/models/research/pcl_rl/model.py", line 135, in setup
    self.actions)
  File "/home/viktor/DeepRL/models/research/pcl_rl/policy.py", line 405, in multi_step
    time_length = tf.shape(all_obs[0])[0]
IndexError: list index out of range

I have TF 1.4 installed with Python 3.5.2. Looks like code was developed on Python 2.7, I had to replace all xrange with range function to be able to start it. And then got an error above.

[drpngx]

Sorry, this is not our code. We can't really support that.

FWIW, For xrange, we usually get it from six.

[ashaw596]

After running it through a 3 to 2 converter. The main remaining bug is you need to convert all 'sampling_dim / 2' in policy.py to 'sampling_dim // 2'. Not 100% sure thats all the bugs. But it runs after that.

[ewanlee]

zip returns an iterator in Python 3.x instead of a list like it did in Python 2.x. This means that, after you unpack it once, it will become exhausted and no longer usable. So there are some bugs in setup_placeholders method in model.py @ofirnachum