Minor Python 3 fixups.

[darrengarvey]

Non-functional changes to run with Python 3.

Incremental printing looks uglier in python 3 unfortunately (print 'foo', => print('foo',end=' ')), but the result should be the same.

[gwern]

ProGAN will also need some updates for CUDA 10/the next Tensorflow version, as they move one of the NCCL functions:

diff --git a/tfutil.py b/tfutil.py
index cf7ad0a..af970ed 100755
--- a/tfutil.py
+++ b/tfutil.py
@@ -13,6 +13,7 @@ import imp
 import numpy as np
 from collections import OrderedDict
 import tensorflow as tf
+from tensorflow.python.ops import nccl_ops
 
 #----------------------------------------------------------------------------
 # Convenience.
@@ -328,7 +329,7 @@ class Optimizer:
                     for var_idx, grad_shape in enumerate(self._grad_shapes):
                         g = [dev_grads[dev][var_idx][0] for dev in devices]
                         if np.prod(grad_shape): # nccl does not support zero-sized tensors
-                            g = tf.contrib.nccl.all_sum(g)
+                            g = nccl_ops.all_sum(g)
                         for dev, gg in zip(devices, g):
                             dev_grads[dev][var_idx] = (gg, dev_grads[dev][var_idx][1])